This article is the second in-depth series. The first article refers to "Automotive Electronics, the Next Apple Industry Chain"

Write in front:

After the era of smart phones, the next ten times the investment opportunities in the electronics industry will appear in the automotive electronics industry. This article focuses on investment opportunities in the automotive electronics-driverless-sensor industry.

1. The core of the sensor is millimeter wave radar and lidar (more important).

2. In the 5G era, the information obtained by millimeter wave radar is transmitted through the C- V2X network.

3. Use accurate maps and other carriers to transfer valuable information to surrounding users.

4. Edge networks may play a key role in the above process.

Abstract

ADAS is a prelude to fully autonomous driving, with a surge in sensor demand.While fully autonomous driving continues to grab the headlines, advanced driver assistance systems (ADAS) have quietly set off a wave of change, fundamentally changing the way traditional cars are handled and the user experience. The redundancy and fault tolerance characteristics of automatic driving require more sensors for higher-level automatic driving. According to our industrial chain research, 2018-2019 is the stage of L2 automatic driving worldwide. It is expected that from 2020, domestic and foreign countries will officially enter the stage of L3 automatic driving. The cross-integration between sensors will greatly increase the demand to ensure the safety of driving as much as possible. Based on the sensors equipped with self-driving cars tested by top self-driving players at home and abroad and the sensors that will be added in the 5G era, the core sensors to be equipped for full self-driving in the future mainly include camera (Camera), millimeter wave radar (RADAR), laser radar (LiDAR), inertial measurement unit (IMU) and vehicle-road coordination system (V2X).

The car camera is the eye of the car, the competitive landscape is relatively concentrated,The global CR4 is 50%, but the strong presence of consumer electronics giants such as Shunyu Optics, which has been fought out in the smartphone industry in China, is expected to officially open the replacement of foreign suppliers with the rapid response of cost performance and localization. Night vision function and AI implantation front-end will be the visible trend of car cameras in the future, and the Chinese market size is expected to be close to 6 billion yuan in 2020.

Millimeter-wave radar, standard for high-end autonomous driving.The global millimeter-wave radar market is highly concentrated, with CR5 as high as 68% in 2018, which is basically monopolized by foreign oligarchs such as Bosch and mainland China. In recent years, independent manufacturers have poured in one after another. Unlike on-board cameras, independent manufacturers are generally small in scale and have little cost-effective advantage in single products. Packaging solutions may be a better breakthrough path. The road traffic environment with Chinese characteristics also provides an opportunity for independent suppliers to provide customized products and services. As the price of CMOS process continues to decline, 77GHz is expected to become the mainstream in the future. It is estimated that the global and Chinese markets are expected to reach 5.12 billion US dollars and 7.21 billion US dollars respectively in 2020.

Lidar can quickly and accurately copy the surrounding 3D environment map with progress up to centimeter level, and is the core sensor to ensure full self-driving with sufficient safety redundancy.The high cost is the main obstacle to lidar assembly, and the 18-year Velodyne announced that its popular 16-line Lidar price has dropped to $3999, which will effectively accelerate the autopilot process, but is still not enough to support the popularity of full autopilot. At present, lidar companies have not yet formed a brand effect and user stickiness, and the technology trend is still in the exploration period. No company has obvious technical advantages and the market is variable. The global market space is expected to exceed $20 billion in 2023.

High-precision positioning sensor: L3 and above are standard for automatic driving.High-precision positioning sensors are standard for high-level autonomous driving. The combination of GNSS(RTK) and IMU is the mainstream solution for high-precision positioning at this stage. It is estimated that by 2025, the market size of GNSS(RTK)& IMU in the vehicle front-end market will exceed $2.5 billion/year.

V2X: Road collaboration to accelerate the landing of the autonomous driving industry.V2X is the "over-the-horizon" sensor for autonomous driving. The V2X infrastructure, the smart road network, is expected to mature rapidly against the backdrop of the domestic government's push for new technology infrastructure. Only calculating the intelligent transformation of highways, the market size of the V2X road end is expected to exceed 130 billion/km, and the potential market size of the V2X vehicle end is expected to exceed 5.6 billion/year.

Key recommend:

Electronic industry:Weir shares, Shunlu Electronics and Wentai Technology;

Computer Industry:Siwei Tuxin, Zhonghai Da and Wanji Technology;

Automotive industry:Desai Siwei and Baolong Technology.

Risk Warning:Macroeconomics is lower than expected, downstream ADAS penetration rate is lower than expected, localization process is lower than expected, new energy vehicle industry chain is lower than expected, etc.

1. Autopilot accelerates, core sensor first

1.1. Autonomous driving core sensors, incremental market space is broad.

U.S. self-driving classification unified industry awareness.According to the SAE standard, automatic driving is divided into 0-5 levels according to the operating authority of human drivers, of which 2-3 levels are an important turning point, marking the formal transfer of driving rights from humans to unmanned systems. The classification of driverless systems provides unified guidance for the driverless research and development process of global car companies.

Global autopilot is accelerating.California is the first region in the world to pass driverless car laws and regulations, and it is also the most important driverless test base in the world. As of January 2018, the California vehicle Administration has issued 50 driverless test licenses. Its annual departure report reflects the technological trend and competitive pattern of global autonomous driving.

2017 was a year of accelerated autopilot, with a record number of new licenses added that year. According to regional distribution, as of January 2018, American enterprises have the largest number of road test licenses, followed by Chinese-funded and Chinese-background enterprises. According to fields, the main participants of automatic driving road test include car enterprises, first-class suppliers of spare parts, Internet companies, start-up companies, and most of the start-up companies that have obtained road test licenses at present, including 13 Chinese-funded and Chinese-background enterprises and 11 American enterprises, chinese innovative companies are the most enthusiastic about autonomous driving.

Google Waymo and General Cruise represent the highest level of autonomous driving for ICT and OEM companies, respectively.The latest "2017 Autopilot" Disengagement Report "released by the California DMV (the 2018 report has not yet been released) includes a total of 20 enterprises, excluding BMW, Ford, Honda, Weilai Auto USA, Volkswagen USA and wheego that have not conducted road tests, Tesla that has not conducted road tests in California, and Faraday Future that has not submitted a report to the California DMV, among the remaining 12 road test car companies, Waymo, owned by Google, ranks first in terms of cumulative road test mileage, with an average manual intervention frequency of 5596 miles. The maturity of self-driving technology is the highest among the 12 companies. General Cruise also has good performance. With a road test course of 131000 miles, the largest number of road test vehicles, the average manual intervention is only once every 1254 miles, judging from this report, Waymo and General Cruise have reached the highest level of ICT and OEM enterprises respectively.

Domestic self-driving road test to step up, Baidu road test mileage far ahead.In December 2017, Beijing issued a policy guidance document for road testing of self-driving vehicles, focusing on breaking through a series of issues related to policies and regulations, management subjects, test sites, test procedures, accident insurance compensation and test supervision for road testing of self-driving vehicles. In 2018, temporary test licenses were issued to 56 self-driving vehicles from 8 enterprises including Baidu, Shanghai Weilai, BAIC New Energy, Pony Zhixing, Daimler, Tencent, Didi Travel and Audi. Among them, Baidu's road test cumulative mileage is far ahead, and only BAIC New Energy is on the list.

From the perspective of autonomous vehicles tested on the road at home and abroad, they are equipped with a variety of sensors, including laser radar (LiDAR), radar (RADAR), camera (Camera) and inertial measurement unit (IMU).

The highest level of self-driving Google Waymo(L5 level) is equipped with sensors such as cameras, millimeter wave radar, lidar, and audio detection systems.Waymo's vision system consists of several sets of high-resolution cameras designed to work in long-range, daylight and low-light conditions; Waymo's millimeter-wave radar uses wavelengths to sense objects and motion. These wavelengths can be used in raindrops and other objects on Friday. Propagation, so that it can work effectively during the day, night, rain and snow, mainly using 77GHz millimeter-wave radar; waymo's lidar emits millions of laser pulses in 360 degrees for ranging, including short-range lidar, high-resolution medium-range lidar, and a new generation of powerful long-range lidar.

General Cruise is equipped with laser radar, millimeter wave radar and camera.According to the General Cruise2018 Safety Report, the General Unmanned Vehicle (L4 level) uses 5 lidars, 21 millimeter wave radars, and 16 cameras. Among them, the camera is to monitor pedestrians, traffic lights, etc., the top laser radar is used to monitor static and dynamic objects, the long-range millimeter wave radar is used to detect vehicles and measure speed, the short-range radar is used to monitor surrounding objects, and the high-resolution radar is used to monitor long-distance moving vehicles.

Tesla is equipped with laser radar, millimeter wave radar and camera.Tesla Autopilot 2.0(L2 level automatic driving) is equipped with 8 cameras, 1 millimeter wave radar and 12 ultrasonic radars. The four side-view cameras in the vision system can basically ensure L3-level functions such as lane change, confluence and high-speed exit. The detection range of the millimeter wave radar on board is 160 meters, and the guess should be 77GHz. Tesla's ultrasonic radar doubles the perceived distance.

The 5G Era,C- V2XIt will be an over-the-horizon sensor for L4/L5 advanced autonomous vehicles. The current autonomous driving solution is mainly to perceive the surrounding environment through radar, cameras, etc., which may be affected by factors such as weather and distance,C- V2XAnother interactive channel for obtaining surrounding environment information is provided. Ma Dejia, senior vice president of Qualcomm Engineering and Technology, pointed out that in the future, with 5G providing greater data capacity for in-car information entertainment and on-board information processing, C- V2X will become an "over-the-horizon sensor" for cars, which can supplement existing sight sensors such as radar, such as prompting drivers to have cars or bicycles approaching in corner blind areas when turning, and providing drivers with signal lamp status information, thus bringing more efficient and safer traffic.

Based on the sensors equipped with self-driving cars tested by top self-driving players at home and abroad and the sensors that will be added in the 5G era, the core sensors to be equipped for full self-driving in the future mainly include laser radar (LiDAR), radar (RADAR), camera (Camera), inertial measurement unit (IMU) and vehicle-road coordination system (V2X).

1.2. Penetration rate: L2 level automatic driving accelerated popularization, penetration rate will increase rapidly.

In 2018, car companies basically realized the L2 automatic assumption, and the future is expected to accelerate the popularization and upgrading.According to the statistics of automotive electronic design and automotive electronic headlines, most foreign and autonomous car owners have realized L2 level automatic driving on some models in 2018. Audi is the leading foreign car company. The A8 launched by Audi in 2018 is the world's first mass-produced L3-level self-driving car. The modified car is equipped with 5 millimeter wave radars, 5 cameras, 12 ultrasonic radars and a 4-line mechanical lidar. The rest of the foreign cars have basically realized L2-level self-driving. Autonomous car companies also realized L2 automatic driving in 2018. According to the plans of most car companies, L3 automatic driving will be realized in 2020.

In addition to traditional car companies, new forces such as Weilai, Xiaopeng and Singularity, which are independent car companies, have also shown positive performance in automatic driving. The L3 automatic driving system of Xiaopeng and Desai Siwei is expected to be implemented in 2020, when it will provide three intelligent functions according to the characteristics of the domestic driving environment, namely, low-speed valet parking, medium-speed traffic jam assisted cruise and high-speed valet driving.

Level 2 autonomous driving penetration is expected to reach more than 30% by 2020.At present, most car companies have implemented L2-level automatic driving on some models, and the sales of models corresponding to production vehicles directly affect the penetration level of automatic driving. According to the 2016 "Made in China 2025", the penetration rate of domestic level 1 and 2 autonomous driving will reach 20% in 2020, the penetration rate of level 3 autonomous driving will reach 10%-20% in 2020, and the penetration rate of level 4 autonomous driving will reach 10% in 2030. In December 2018, the Ministry of Industry and Information Technology released the "Vehicle Networking (Intelligent Networked Vehicle) Industry Development Action Plan" to raise the L2 penetration target from the target of 20% penetration rate of Level 2 automatic driving to more than 30% penetration rate by 2020, and L2 automatic driving to accelerate penetration.

1.3. Bicycle demand: ADAS continues to upgrade, bicycle demand continues to rise

As the level of autonomous driving is upgraded, the demand for sensors in bicycles continues to rise.Take Tesla, Audi A8 and General Cruise as examples. The number of cameras is basically increasing with the increase of the number of levels. The increase of millimeter wave radar is especially obvious with the increase of levels. The number of millimeter wave radars carried by L4 General Cruise is 4 times that of Audi A8. The expensive laser radar is necessary at L3 level, and the demand for L4 level is also significantly increased, the number of L5-class Waymo cameras and millimeter-wave mounts, the most mature technology, has decreased, but the number of high-value lidar mounts is increasing.

The penetration rate of autonomous driving is gradually increasing, and the demand for sensors is expected to grow exponentially, with the market size expected to reach $3.1 billion in 2022 and $77.3 billion in 2032.According to Maims Consulting, the global output of self-driving cars is hundreds in 2017, and the global output of self-driving cars is expected to reach 23.1 million in 2032, with a CAGR of 58%. Overall revenue associated with autonomous vehicle production will be $300 billion billion, with 52 percent coming from the vehicle itself, 26 percent from sensor hardware, 17 percent from computing hardware, and the remaining 5 percent from integration. Lidar, radar, cameras, inertial measurement units and global navigation satellite systems are all growing exponentially, with a combined market size expected to reach $3.1 billion in 2022 and $77.3 billion in 2032.

2. Car camera: independent suppliers accelerate the entry to open the process of domestic substitution.

2.1. Composition and classification of vehicle-mounted cameras

On-board vision sensors, the new blue ocean of cameras.The car camera is an important sensor for automatic driving, mainly including lenses, filters, CMOS, PCBA, DSP and other packaging and protection materials. Different from the mobile phone camera, the module process of the car camera is much more difficult, mainly because the car camera needs to maintain a stable working state for a long time under various complex working conditions such as high and low temperature, humidity, strong light and vibration. On the whole, the technical barrier of the car camera is significantly higher than that of the mobile phone camera. With the rise of automatic driving, the car camera has become the new blue ocean of the camera.

There are many kinds of on-board cameras, and the loading volume is increasing.Vehicle cameras play an irreplaceable role in the realization of autonomous driving, such as identifying traffic signs, traffic lights and pedestrians. Take Tesla Model 3 as an example. A total of 9 cameras are installed inside and outside the car, including 3 front-view cameras, 2 side-view cameras, 2 rear-view cameras, 1 rear-view camera and 1 internal-view camera. Among them, the front-facing camera uses a three-eye camera, mainly because the monocular camera has an irreconcilable contradiction in the range and distance of ranging, that is, the wider the viewing angle of the camera, the shorter the length of the precise distance that can be detected, the narrower the viewing angle, and the longer the detected distance. The three cameras are narrow-angle cameras (35-degree field of view, maximum distance 250 meters), medium-range camera (50 degree field of view, maximum distance 80 meters) and fisheye camera (150 degree field of view, maximum distance 60 meters), left and right side-view wide-angle camera (80 degree field of view, maximum distance 60 meters), left and right rear-view medium-range camera (60 degree field of view, maximum distance 100 meters), fisheye rear-view camera (140 degree field of view, the maximum distance is 50 meters), the inside view camera monitors the driver's attention.

2.2. Independent suppliers accelerate the entry, domestic substitution is imminent.

The global car camera CR4 is 50%, the competitive landscape is relatively concentrated, and independent suppliers are beginning to enter. Compared with consumer electronics and industrial electronics for industrial vision such as mobile phone cameras, vehicle cameras have higher requirements for stability and specifications due to factors such as safety and use environment, complex module packaging process, high technical barriers, especially the dispensing process. The barriers of car cameras mainly lie in module packaging and customer barriers. At present, Panasonic and Sony occupy a large market share in the world, but the overall competition pattern is not very concentrated. In addition to Panasonic's 20% market share, the market share of the following eight major competitors is not far from each other in ChinaShunyu Optics, Ophelia Light, Desai Siweiand so on have begun to fully enter the car camera module package manufacturing.

(Dry goods research note:There may be big opportunities in this, similar to the investment opportunities of 10 times shares brought by Apple's industrial chain)

Foreign suppliers occupy the mainstream position, independent suppliers accelerate the entry, domestic substitution process started.International multinational giants such as Bosch, Vining, Continental Tamik and Gele occupy the mainstream joint venture brand supply chain by virtue of their technological advantages and customer support advantages, but their products are mainly camera-based. Autonomy from rear to front, from commercial vehicles to passenger cars, from autonomy to joint ventures, from single customer to multi-customer breakthroughs, although there is still a gap between technology and foreign capital in the short term, but the operation is flexible, from car cameras to specific applications, such as 360 surround view, automatic parking, driving recorder, vehicle-mounted monitoring video recorder system, the product is very rich, can fully meet the needs of the main engine factory customers and after-sales market. In the future, with the continuous maturity of independent supplier technology and the increasing scale, with the traditional advantages of rapid response and cost performance, it is expected to open the substitution of foreign suppliers.

2.3. Future trend and space of vehicle camera

One of the future trends: night vision camera or become the standard car camera.According to the statistics of the National Highway Traffic Safety Administration of the United States, although night travel accounts for only a quarter of the road traffic in the United States, traffic accidents account for half of the country, mainly due to poor vision at night. This requires that the vehicle camera must have a strong photosensitive capacity, and the future night vision system may become the standard of the vehicle camera. From the actual effect, cameras with night vision function can greatly improve driving safety. For example, Tesla's Autopilot HW2.0 camera uses four filters-RGGB (red, green, green, blue) to create a single color pixel on the cell (two green colors are used to improve resolution/brightness). Three of the camera's filters are monochromatic visible light, and the other uses a red filter (RCCC) to increase the sensitivity of monochromatic light and detect red traffic lights and taillights.

The second trend of the future: AI algorithm and AI chip implanted camera hardware front-end is an important trend in the future.At present, many artificial intelligence automobile start-ups at home and abroad are trying to implant AI into the hardware front end of the camera to develop artificial intelligence cameras with target detection, segmentation and recognition capabilities, and even parameter estimation and behavioral intention prediction functions. For example, Utility announced in October 2018 that it would cooperate with Sony to jointly develop the "intelligent patrol" on-board camera influence system. By using AI technology to identify the license plate number, law enforcement personnel no longer have to manually enter the license plate number, thus avoiding the distraction of law enforcement vehicle drivers as much as possible.

Car camera loading volume is rising rapidly, and the market size is expected to be nearly 6 billion yuan in 2020.According to statistics from Zos Production and Research, 6.39 million cameras were loaded in the domestic passenger car market in 2017, mainly applied to reversing images (rear view) and 360-degree panoramic cameras (surround view). Front view applied to FCW, LDW, AEB and other functions and internal view monitored by drivers are the main growth points of vehicle cameras in the future. High-tech intelligent vehicles predict that the penetration rate of rear view cameras is expected to reach 50% and that of front view cameras is expected to reach 30% in 2020. With the improvement of the maturity of the industrial chain, the price of on-board cameras has also continued to decline, from more than 300 yuan in 2010 to 150 yuan in 2018, and the price of general blind spot cameras has basically dropped to less than 100 yuan. According to the forecast of high-tech intelligent vehicles, the market size of domestic vehicle cameras is expected to be close to 6 billion yuan by 2020.

2.4. Changes in Tesla's self-driving vision program.

Two phases of Tesla's self-driving technology program.Tesla initially adopted the Mobileye monocular camera scheme. After many traffic accidents, Tesla changed to the technical scheme of millimeter wave radar as the main part and camera as the auxiliary part. The number of cameras was also expanded to 8 at one stroke, corresponding to the Autopilot 1.0 system and the Autopilot 2.0 system respectively. AutoPilot1.0 system has 1 front camera, 2 front and rear radars and 12 ultrasonic sensors on the hardware, while 2.0 system not only enhances the radar and increases the distance of the sensors on the hardware, but also provides 3 front cameras with different focal lengths, 2 side cameras and 3 rear cameras. The number of cameras is expanded from 1 to 8 in the 1.0 system.

The monocular camera scheme is mature in the short term, but it is difficult to achieve L5 level automatic driving.On May 7, 2016, a Tesla Model S, which turned on its automatic driving, crashed into a large white truck turning sideways. According to the results of the accident investigation, the reason was that the vehicle image recognition system failed to distinguish the white compartment of the truck from the blue sky and white clouds behind it. This is mainly due to the fact that Tesla adopted a Mobileye monocular camera technology scheme at that time. Its identification of vehicles was carried out by means of feature points, such as tail lights and rear wheels to identify the front car. It lacked effective response to how to distinguish cars, white clouds or lake surfaces in more complex real-life scenes. The monocular vision detection scheme is more suitable for ADAS or low-level automatic driving. Tesla decisively chose to turn when he realized that the Mobileye monocular camera scheme could only dominate in the short term and could not ultimately support its goal of fully automatic driving, and Tesla's automatic driving route then entered the 2.0 era.

The advantages and disadvantages of the three-eye camera scheme are significant and remain to be tested by time.The three-eye cameras adopted by Tesla are respectively front-view narrow-view camera, front-view main-view camera and front-view wide-view camera, which realize the balance of field of view and distance. According to their different focal lengths, they are respectively responsible for long-distance ranging, medium-distance ranging and near-distance ranging (including identification of traffic lights, road obstacles, etc.). The three cameras overcome the limitations of monocular and binocular vision, ranging accuracy, installation position and other aspects of the hard injury, the maximum simulation of human eyes fast zoom, while covering the long and short distance range characteristics. However, with the increase in the number of cameras, the accuracy error rate has also increased. Sanmu adopts the method of calculating the estimated distance of parallax in real time. It needs to calculate the distance, direction and speed information of all dynamic and static objects at different times at the same time, and the calculation amount soars. In addition, the three-way data are transmitted synchronously, but the collected data are not always the same, and the system background algorithm does not have certain operation logic to review, resulting in the accuracy of driving decision-making cannot be estimated. Considering the cost, reliability and accuracy of the three-eye camera solution, whether it is the best solution to achieve fully autonomous driving remains to be tested by time.

3. Millimeter-wave radar: All-weather sensor, high-order automatic driving standard

3.1. Principle and composition of millimeter wave radar

Millimeter wave radar three steps, environmental perception, calculation analysis, control execution.The vehicle-mounted millimeter-wave radar transmits millimeter-wave through the antenna, receives the target reflection signal, quickly and accurately obtains the physical environment information around the car body (such as the relative distance, relative speed, angle, movement direction, etc. between the car and other objects) after rear processing, and then tracks, identifies and classifies the target according to the detected object information, and then combines the body dynamic information for data fusion, the intelligent processing is finally carried out by the central processing unit. After a reasonable decision, the driver is informed or warned by various methods such as sound, light and touch, or active intervention in the car in time, so as to ensure the safety and comfort of the driving process and reduce the probability of accidents.

Millimeter-wave radar is an all-weather all-day ADAS sensor and standard for high-end automatic driving.The mainstream products of millimeter wave radar are 24GHz and 77GHz. The former is mainly used for short-distance sensing and is placed at the rear and side of the car. It can detect the surrounding environment and blind spots of the car body and realize parking assistance, lane change assistance and other functions. The latter is used for long-distance measurement and is mainly placed in the front of the car to realize automatic car following, adaptive cruise, emergency braking and other functions. Millimeter wave radar system mainly includes antenna, transceiver module, signal processing module, etc. Among them, the key component is front-end monolithic microwave integrated circuit (MMIC), which includes a variety of functional circuits, such as low noise amplifier, power amplifier, mixer, etc. MMIC can simplify the structure of radar system, facilitate large-scale production, reduce the logistics cost of the system, and accelerate the application of millimeter wave radar.

3.2. Foreign oligopoly, domestic capital began to break the situation.

The global millimeter wave radar market is highly concentrated, with CR5 as high as 68% in 2018 and a foreign oligopoly.From the perspective of the competitive landscape, the global millimeter-wave radar market is basically monopolized by foreign auto parts giants led by Bosch, of which Bosch is dominated by 77GHz millimeter-wave radar. At present, China's 24GHz vehicle radar market is mainly dominated by Valeo, Hella and Bosch, accounting for more than 60% of the total shipments. China's 77GHz on-board radar is mainly dominated by Continental, Bosch and Delphi, accounting for more than 80% of total shipments. On the whole, the Chinese market is still under the monopoly of foreign oligarchs.

Domestic manufacturers into the millimeter wave market, customized solutions or domestic alternative breakthrough.The domestic millimeter wave radar market is currently monopolized by foreign parts giants such as mainland China, Delphi and Bosch, and is constantly introducing newer and better products. When the performance and price of a single product are not enough to open the gap with foreign suppliers, packaging solutions is a better breakthrough path, including how much redundancy the whole system has and providing customized services. According to a survey report of a joint-venture main engine factory, more than 90% of consumers shut down LDW and FCW functions because of poor functions. A typical scene with Chinese characteristics is merging and cutting, because domestic cutting-in behavior is more frequent than foreign ones, and localized driving assistance systems should be more sensitive to the cutting-in of adjacent lanes. Foreign ACC systems usually wait until the cars cutting into adjacent lanes merge and straighten the rear of the cars, which is prone to rear-end collision. Therefore, the more effective way for independent manufacturers to replace foreign manufacturers is to develop customized solutions, together with cameras, to do multi-sensor fusion driving and auxiliary applications.

3.3. Product upgrade, 10 billion market space to be explored

77GHz on-board millimeter-wave radar is the mainstream of the future, and the gap with 24GHz radar shipments has been significantly narrowed in 2018.As autonomous driving moves from L2 to L5, the number of millimeter-wave radars required by cars is increasing, which requires smaller, lower power and lower price millimeter-wave radars. The biggest advantage of 77GHz is that the antenna is a 24GHz 1/3. The same volume can do more channels, the recognition accuracy is higher, and the penetration ability is stronger. The challenge is mainly in terms of design and price. From an industry perspective, DIGITIMES Research estimates that the price of high-frequency millimeter wave radar will fall by 50% in 2022 compared with 2017 after the introduction of CMOS process, and the decline in 77GHz price is expected to accelerate its popularity. Judging from the loading volume of on-board millimeter-wave radar in 17-18 years, it accelerated significantly in 17 years, with a year-on-year increase of 104.6 percent. In 18 years, affected by the decline in car market sales, there was still a year-on-year growth of 54%. According to Zoss production and research data, the 77GHz radar achieved an anti-surpass of the 24GHz radar in December 2018, one year earlier than originally expected, and the upgrade process of the on-board millimeter wave radar exceeded expectations.

With the accelerated popularity of autonomous driving, the global and Chinese automotive millimeter wave radar market is expected to reach $5.12 billion and $7.21 billion, respectively, in 2020.With the increasing penetration of ADAS, the market demand for vehicle millimeter wave radar is increasing. According to statistics compiled by the China Business Industry Research Institute, the market size of vehicle-mounted millimeter wave radar in the world and China in 2015 was about US $1.94 billion and 1.8 billion yuan respectively. Considering that the emergency automatic braking system (AEB), which must carry at least one millimeter wave radar, is expected to reach US $5.12 billion and 7.21 billion yuan respectively by 2020.

4. Lidar: core component of autonomous driving

4.1. LiDAR -- Core Sensor of High Precision Positioning System

Accuracy and ranging.Laser radar has shorter wavelength, horizontal resolution can be controlled within 0.1 °, and has strong anti-interference ability and longer detection distance. Therefore, compared with other sensors such as millimeter wave radar, it can provide more accurate and stable positioning and navigation.

Transmitter-receiver-photoelectric converter-signal processor four modules.When working, the transmitter emits a pulsed laser. After the laser encounters an object, it will be diffusely reflected and received by the receiver, and then recognized by the photoelectric detector and converted into an electrical signal to be transmitted to the signal processing module for analysis.

Multi-beam laser enhances positioning performance.The single-line lidar has a transmitter and a receiver. The transmitter generates a frame of data every time it rotates a certain small angle in the radar. Single-line radar has low cost, but can only obtain linear information. Multi-line radar uses multiple transmitters to poll, and one frame of point cloud data can be obtained in one polling cycle. The greater the number of lines, the greater the cloud of points collected in seconds.

Solid-state lidar technology is in the ascendant.At present, the laser radar mainly has two kinds of mechanical and solid state. The mechanical lidar has a large volume and high measurement accuracy. It is generally placed outside the car and is more expensive. Solid-state lidar has a relatively compact structure, low price, and poor measurement accuracy, and can be hidden in the car body. Earlier companies generally used expensive mechanical lidar, represented by Velodyne. Mechanical type means that mechanical devices are used to drive the transmitter to rotate and pitch during scanning, with high accuracy, but the technology is complex and the cost remains high. With the development of technology, solid-state lidar has evolved, which is divided into MEMS, OPA(Optical Phased Array) and Flash. MEMS type uses MEMS micro-galvanometer to integrate all mechanical components into a single chip and then use semiconductor technology to produce it. It is essentially a mixed solid state. OPA type uses the principle of coherence to form an array of multiple light sources, forms a main beam in a specific direction by controlling the light emission time difference of each light source, and realizes scanning in different directions by controlling the main beam. Flash type is a non-scanning radar, which emits surface array light, the reflected signal is received by a high-precision sensor, and a two-dimensional or three-dimensional image is drawn as the output surrounding environment image. Mechanical rotating contrast solid-state scanning, high precision but difficult production process, high cost, more suitable for military-grade, enterprise-class equipment. In pure solid state/mixed solid state, MEMS is small in size, low in cost, low in consumption, and suitable for medium and long distances; Flash is suitable for close use; OPA-type technology is not yet mature, but the market is optimistic about its performance indicators.

Lidar and other sensors have their own advantages and disadvantages, and can be used together to adapt to multi-scene applications.Laser radar has high resolution and strong anti-electromagnetic interference ability, but it is greatly affected by rain, fog and haze, and its detection ability in the near field (<30m) is weak. Therefore, in different scenarios, different sensors have their own advantages, and only when used together can high-precision positioning be achieved in various scenarios.

4.2. From 3D mapping/mapping to autonomous driving.

In the era of autonomous driving, lidar has two main applications. First, it is used for mapping 3D terrain and providing high-precision map database for automobiles. The second is used for vehicle-mounted terminals, which are used in combination with cameras, millimeter wave radar, GPS, etc. to realize centimeter-level accuracy for route navigation.

Autopilot did not arrive, high-precision map first.The daily navigation map belongs to the "car entertainment system", which is far from accurate enough for self-driving cars. Therefore, professional equipment is required to collect detailed geographic terrain data, accurate to the level of street lamps, guardrails and street trees, and built into the car system. In addition to accuracy, 3D map mapping also needs to ensure continuity and timeliness. Continuity means that the map connection must be smooth, and timeliness means that the map must be updated on time to avoid changes in road conditions. The large-scale demand for lidar for mapping will appear in the L4 level of automatic driving.

Lidar is the main sensor of high precision map measurement.Lidar can compensate for the lack of resolution of millimeter wave radar, the lack of all-weather capability of the camera, and the regional problems of GPS navigation. The collection of high-precision map information has a greater demand for information than ordinary maps, so driving sections need to be measured by data acquisition vehicles equipped with lidar. At present, the mainstream solution for high-precision map acquisition is that multiple lidars are distributed around and on the roof of the map data acquisition vehicle, with cheap low-line LiDAR around and more expensive high-line LiDAR on the roof to ensure performance requirements and reduce the cost of the vehicle. For example, the four-dimensional map new data acquisition vehicle can collect panoramic image data, road image data, sub-meter high-precision GPS data, road measurement close-range lidar and other high-precision data.

4.3. Market Space and Large Factory Pattern

Mapping and navigation, the demand for lidar in the era of autonomous driving is huge.

The future growth point of lidar lies in mapping and driving navigation.In the military, meteorological and other fields, lidar has long been widely used, but there is little room for incremental demand. L3 level autonomous driving already requires lidar to provide accurate navigation, and with the full spread of autonomous driving, lidar will fully penetrate the consumer sector.

The supply side is far from mature, opportunities and challenges coexist.

The three foreign giants actively improve the product line, the layout of the car LiDAR market.Globally, Velodyne, Quanergy and Waymo are industry leaders, focusing on reducing the cost of lidar to achieve large-scale civilian use. The price of Velodyne's new 16-line Lidar has been reduced to US $3999. According to media information, Quanergy and waymo respectively announced that when their new products are put into production, the prices will be around US $250 and US $500. At that time, lidar will be fully popularized in cars of all prices.

The technology of domestic companies needs to be further accumulated.China started late in the field of laser radar, and there is a certain gap between the technical level and the world's leading enterprises. The boss company engaged in laser radar research and development and production of the main Guorui technology, four-creation electronics, superstar technology, etc., of which Guorui technology and four-creation electronics to military radar-based, superstar technology developed civilian laser radar products.

Vehicle lidar manufacturers seek to reduce costs as much as possible while ensuring basic performance.The biggest obstacle for lidar to be used in private cars is the high cost. The price of many lidars even exceeds that of the whole car, which is unacceptable to car companies. Map mapping is an enterprise-level application and is relatively insensitive to price. However, considering that high-precision maps must be updated in time in the future, the huge stock gap also puts forward higher requirements for the price of enterprise-level lidar. Different from other mature industries, at present, lidar companies have not yet formed a brand effect, user stickiness, and the technology trend is still in the exploration period. No company has obvious technological advantages, and the market is variable. In terms of demand, the era of automatic driving is accelerating, the development of other subsystems is relatively perfect, and the industrialization of lidar is relatively backward, which has become one of the hardware bottlenecks of the whole unmanned driving.

5. High-precision positioning sensor: L3 and above standard for automatic driving

5.1. Definition of high precision positioning sensor

High-precision positioning sensors are standard for high-level autonomous driving.In order to realize the automatic driving of vehicles, it is necessary to solve the problems of where (immediate position) and where to go (target position). Therefore, high-precision positioning sensors (centimeter-level accuracy) need to be applied to L3 and above automatic driving.

According to the different positioning technology, high precision positioning can be divided into three categories.The first category is signal-based positioning, such as GNSS (Global Navigation Satellite System) positioning; The second category is dead reckoning, which relies on IMU (Inertial Measurement Unit) to infer the current position and orientation according to the position and orientation of the previous moment. The third category is environmental feature matching, which uses laser radar-based positioning to match the observed features with the features in the database and the stored features, get the current car position and attitude. Observing the current high-precision positioning schemes in the industry, they generally take the form of fusion. Generally speaking, there are: 1) sensor fusion based on GNSS(RTK) and IMU; 2) matching of lidar point cloud with high-precision map; 3) road feature recognition based on computer vision technology, GPS positioning is an auxiliary form. Among them, the combination of GNSS(RTK) and IMU is the mainstream solution at this stage.

The fundamentals of the GNSS(RTK)& IMU combination.GNSS in the complex dynamic environment, especially in large cities, the problem of multi-path reflection is very significant, resulting in the acquisition of satellite positioning information is easy to produce a few meters of error. In addition, because the GNSS update frequency is low (10Hz), it is difficult to provide accurate real-time positioning when the vehicle is traveling fast. Therefore, GNSS is usually assisted by inertial sensors (IMU) to enhance the accuracy of positioning. The IMU is a sensor that can detect acceleration and rotation, and the motion trajectory and real-time position information of the body can be obtained by measuring the acceleration and rotation angle. The advantage of the IMU is that it is completely independent, neither restricted by movement nor by any particular environment or location. The last line of defense can be formed when the GNSS signal is lost or fails ".

5.2. High-precision positioning sensor market size

High-precision positioning sensors are standard for L3 and above automatic driving.Referring to the industry in-depth report "Autopilot: The" iPhone "Moment of the 100-Year Auto Industry" released earlier, the GNSS(RTK)-IMU combination is the core component of high-level autopilot (L3 and above), and each autopilot vehicle will come standard with at least one set. According to the industry chain research, the GNSS(RTK)-IMU combination has a high mass production price at this stage. With large-scale mass production and mature technology, the overall price is expected to drop to US $500/piece in 2025. Combining domestic autopilot penetration and price curve forecasts, we estimate that the market size of GNSS(RTK)& IMU in the on-board front-loading market will exceed $2.5 billion/year by 2025.

5.3. High-precision positioning sensor industry pattern

The core of the GNSS(RTK) industry competition is "infrastructure".The positioning accuracy of GNSS satellite signals is at the meter level, which is far from the requirements of automatic driving. In order to achieve centimeter-level satellite positioning, GNSS correction is required to correct positioning errors caused by ionosphere. RTK(Real Time Kinematic Carrier Phase Difference) technology is usually adopted. The calculation terminal equipment at the vehicle end needs to cooperate with "infrastructure" to achieve the best positioning effect. "Infrastructure" can be divided into two categories: 1) "Ground-based enhancement" mode is to correct the error of satellite positioning measurement by establishing a large number of fixed reference stations (CORS stations) on the ground, which is typically represented by Qianxun position; 2) "Satellite-based enhancement" mode is to upload the correction parameters obtained from the reference station to the satellite, and then to the global broadcast through the satellite to correct the error. In essence, the core of the GNSS(RTK) industry competition is not on the "infrastructure" side. Taking the domestic high-precision positioning service leader Qianxing position as an example, it has completed the construction of more than 2200 ground-based enhancement stations in the country, and the huge ground-based enhancement station has become an important competitive barrier to the high-precision positioning service it provides.

The core of IMU industry competition is to balance high precision and low cost.The price of IMU is proportional to the accuracy. Autonomous driving also requires higher accuracy of sensors, but higher accuracy means higher costs. If you can create high-precision, low-cost IMU solutions, you can occupy the technical commanding heights to break technical barriers. At present, the major IMU providers in the world include ADI, Honeywell, Northrop Grumman (the main supplier of Nova Thai IMU modules), Fairchild, Xsens, Sensonor AS, KVH, Applanix (acquired by Tianbao), Epson (one of the suppliers of Nova Thai IMU modules), and Xinna Sensor (ACEINNA) spun off from US-based New Semiconductor Company (MEMSIC) in 2017.

6.V2X: Vehicle-road collaboration to accelerate the landing of the autonomous driving industry.

6.1.V2X definition: "over-the-horizon" sensors for autonomous driving

V2X-equipping automated driving with "beyond visual range" sensors.V2X, as the name suggests, is vehicle-to-everything. The definition of V2X in the industry is to realize information interaction between the vehicle and all entities that may affect the vehicle, and to complete the sensing work through radio wave propagation, that is, wireless communication, so as to reduce accidents, slow down traffic congestion, reduce environmental pollution, and provide other information services. The main application scenarios of V2X include vehicle-to-vehicle (V2V) and vehicle-to-infrastructure (V2I). Simply put, V2X can be understood as a set of sensor redundancy that can see farther than the human eye and is not affected by the weather. It can let the vehicle know the surrounding movement in real time, detect the change of traffic flow outside the line of sight, and send alarm message to the driver. At present, the mainstream sensors equipped with self-driving cars, including cameras, millimeter wave radar, lidar, etc., are basically "nearsighted" (the current detection distance of mainstream sensors is 200-300 meters). The advantage of V2X is that with the help of low-latency, highly reliable network connection interaction, it can allow vehicles to understand the surrounding trends in real time, detect changes in traffic flow outside the line of sight, and send alarm messages to the driver, and there is no need to worry about road conditions and working conditions. The impact of the sensor.

The V2X industry needs a three-pronged approach of "car", "road" and "network" to form an industrial pattern of co-evolution of vehicle and road.If V2X technology wants to play the perfect effect, it needs a three-pronged approach. The vehicle end, road test and communication link all need to be laid out accordingly. Among them, 1) road end upgrade: that is, the intelligent transformation of the road (RSU roadside unit), including the upgrade of basic elements such as road signal lights, electronic signs, cameras, etc. 2) Network upgrade: vehicle-vehicle communication and vehicle-road coordination supported by V2X technology all rely on low-delay and highly reliable network connection interaction; 3) Vehicle-end upgrade: namely, vehicle-end networking transformation (OBU vehicle-mounted unit). The vehicle terminal can integrate functions such as V2X communication, algorithm decision, and APP terminal display.

6.2. Science and technology new infrastructure power intelligent road network, vehicle road collaboration to help self-driving.

Vehicle-road collaboration with V2X as the core is the route of self-driving technology advocated in China.

Vehicle road collaboration with V2X as the core is the technical route for the self-driving industry advocated by the domestic government.Compared with the "bicycle intelligence" self-driving technology route promoted by B- end enterprises such as car factories, Tier1 and Internet technology giants abroad, the top-level design of self-driving of the domestic government advocates the "intelligent network connection" technology route, including "intelligent transformation of bicycles" and "co-evolution of vehicles and roads". By promoting the infrastructure construction of intelligent road network, the automatic driving industry will be accelerated. The "Action Plan for the Development of the Vehicle Networking Industry" issued by the Ministry of Industry and Information Technology in 2018 clearly defines the industry's goal: to achieve LTE- V2X coverage on some expressways and major urban roads by 2020, to carry out 5G-V2X demonstration applications, to build a vehicle-road collaborative environment, and to achieve a high degree of "human-vehicle-road-cloud" collaboration.

The domestic government is pushing hard for new technology infrastructure, and V2X's infrastructure, the smart road network, is expected to mature rapidly.For 19 years, the government has put forward a policy development approach for a new "infrastructure" for science and technology. Specifically, Lian Weiliang, deputy director of the National Development and Reform Commission, said that this year will increase investment around the two keywords of "construction and transformation. "Construction" focuses on five aspects, "strengthening the construction of new infrastructure" in the first place, including promoting the construction of artificial intelligence, industrial Internet, Internet of things, etc., and accelerating the pace of 5G business. In addition, Miao Wei, Minister of the Ministry of industry and information technology, also made it clear in an interview with CCTV news that 5g temporary licenses will be issued in several cities this year, taking the lead in realizing large-scale networking in hot spots; at the same time, the pace of network construction such as base stations will be accelerated to gradually cover the whole country. We should accelerate the maturity of the terminal industry and promote the application of 5G in more fields such as car networking. We expect to tilt the government's financial support with industrial policy support, is expected to quickly improve the new science and technology "infrastructure", as soon as possible to achieve the transformation of intelligent road network, paving the way for the self-driving industry.

China is expected to have the world's first-class V2X industry infrastructure

Road end: domestic funds available for intelligent road construction are abundant.The domestic government's investment in the highway field is much higher than that of developed countries in Europe, America and Japan, and the funds available for intelligent road construction are also more abundant. At the same time, domestic government departments have also issued relevant policies for smart highway construction, providing guidance and basis for roadside intelligent construction investment from the top-level design.

Network side: China's 5G network deployment is the world's leading.According to Ernst & Young, 5G has been a high priority on China's national agenda. Under the guidance of national-level strategies such as the Made in China 2025 and the 13th Five-Year Plan, government departments have formulated supportive policies. China's 5G technology development is the world's largest government-planned 5G program. China is ready to launch commercial 5G services in 2019, a year ahead of schedule. China will join countries such as the United States, South Korea, Australia and the United Kingdom as the world's first market to distribute 5G services.

6.3.V2X Industry Scale & Commercial Progress

V2X industry progress: communication standards established, commercial products mature

V2X communication standards established. The domestic V2X communication standard is C- V2X.Among them, C in the C- V2X refers to cellular (Cellular), which is a wireless communication technology for vehicles based on the evolution of cellular network communication technologies such as 3G/4G/5G. It includes two communication interfaces: one is a short-range direct communication interface (PC5) between vehicles, people and roads, and the other is a communication interface (Uu) between terminals and base stations, reliable communication over long distances and greater ranges is possible. C- V2X is a communication technology based on the 3GPP global unified standard, including LTE-V2X and 5G-V2X. From the perspective of technology evolution, the LTE-V2X supports the smooth evolution to 5G-V2X. Compared with other V2X technologies (Drsc), C- V2X has excellent performance and cost-effectiveness, and can also be forward compatible with 5G, these factors make C- V2X direct communication the preferred solution in China. At the same time, C- V2X is currently the only V2X technology that follows the global 3GPP standard and supports continuous evolution to achieve 5G forward compatibility; it has received extensive support from the global automotive ecosystem including the fast-growing 5G Automotive Alliance.

C- V2X is expected to carry out pre-commercial testing in 2019 and large-scale commercial use in 2020.According to the "C-V2X White Paper" released by CAICT of China Information and Communication Research Institute in 2018, according to the progress of industrial development, C- V2X technology will be tested in stages: before 2019, industrial forces will be concentrated to promote LTE-V2X technology tests and product maturity; 5G-V2X Uu technology test will be carried out in 2019. More specifically, the LTE-V2X will start large-scale tests in June 2018 to upgrade and transform roadside infrastructure and verify the networking performance of the network and typical vehicle networking service performance under multi-user conditions. In 2019, some city-level infrastructure will be renovated and pre-commercial tests will be carried out. In 2020, we will promote the commercial use of LTE-V2X and support the commercial application of intelligent travel services such as traffic efficiency.

V2X commercial deployment supporting products mature.According to the industry chain research feedback, the current V2X commercial deployment of supporting products after pre-research and a large number of testing, has entered the mass production state, can cooperate with the "network end" construction, to achieve rapid commercial deployment. Among them, 1) road end: giants including and others have released commercial RSU (roadside unit) products in 2018. 2) Vehicle end: With the technical support of Qualcomm and other chip factories, the major domestic mainstream communication module suppliers, vehicle terminal (OBU) equipment providers have released commercial products, ready for the large-scale deployment of the industry.

V2X Market Size Estimation

V2X market size estimates.At present, the total mileage of highways in the country is 4.34 million kilometers, of which there are about 136000 kilometers of expressways (2017 data). According to the industry chain research, the cost of intelligent transformation of highways at this stage is expected to be around 50-1 million/km. Only calculating the intelligent transformation of highways, the market size of the V2X road end is expected to exceed 130 billion/km. In addition, according to the industry chain research, the production price of the basic communication module at the V2X road end is expected to be 100-200 yuan/vehicle. According to the domestic sales volume of 28.146 million vehicles in 2018, the potential market size at the V2X vehicle end is expected to exceed 5.6 billion/year.

V2X market industry map.1) Road end: road end transformation is expected to be led by government investment, intelligent transportation information manufacturers are expected to cut into the road intelligent transformation of the big market; 2) network end: the upgrade of communication network will be led by the input of communication operators, the relevant communication equipment manufacturers are expected to benefit; 3) car end: it is expected that the car factory will lead the upgrade of the car end. Car companies determine the commercial timetable for car upgrades.

7. The only way for automatic driving, multi-sensor fusion

7.1.360 Surround View Enters the Era of Full Popularization

Multi-camera fusion, 360 surround view system.The 360 surround view system (AVM) is composed of multiple cameras around the vehicle body, image acquisition components, video synthesis/processing components, digital image processing components, and on-board displays. The cameras around the car body take images of the front, rear, left, and right of the car respectively, usually 4 cameras. The images are converted into digital information by the image acquisition component and sent to the video synthesis/processing component. The image processed by the video synthesis/processing component is processed by the digital image processing component. After conversion to analog signal output, the panoramic image information of the car and its surrounding environment is generated on the on-board display installed inside the car. The core difficulty of the technology is how to accurately fit the pictures taken by the four cameras together seamlessly without distortion, which requires high enterprise algorithms and belongs to the fusion of multiple cameras.

The penetration rate of 360 is rapidly increasing and entering the era of full popularization.Since 2014, China's auto market has fallen into low growth, but the number of products has increased unabated, and competition has become fierce. Reversing and warehousing is one of the core pain points of consumers' daily car use. Benefiting from the continuous decline in costs, in the fierce competition, some manufacturers have begun to assemble 360 surround view systems to get on the car as an important selling point. According to smart car statistics, the comprehensive configuration rate has increased year by year, and increased significantly in 2017 and 2018. In 2018, the standard rate of panoramic cameras reached 20.42 percent, and the matching rate also reached 6.23 percent.

360 surround view system, joint venture assembly rate is generally not high, independent differentiation is obvious, luxury car assembly rate is the highest.We have counted the situation of the mainstream models in the market equipped with 360, which are divided into three groups: joint venture, autonomous and luxury cars, which are very different from each other. In terms of joint venture vehicles, only Accord and Passat's high-end models are equipped with 360 surround view systems. In terms of autonomy, the overall configuration rate of 360 surround view system is significantly higher than that of joint venture vehicles, but the internal differentiation is obvious. The assembly rate of independent high-end vehicles such as Great Wall Haver H6, Geely Boyue, Chuanqi GS5, BYD Tang and SAIC MG HS is generally higher than 40%, however, low-end models such as Geely Emgrand, Baojun 510, Changan CS35 and Rongwei RX3 are not equipped with 360 surround view systems. In terms of luxury cars, the overall assembly rate is the highest, but the differentiation is also obvious. The assembly rate of luxury brand mid-range models, such as Mercedes-Benz E-Class, BMW 5 Series, BMW X3 and Audi A6L, has reached 100, however, entry-level models such as BMW X1, 1 Series and Audi A3, Q2L, Q3 are not equipped with 360 surround view systems.

7.2. Fully automatic parking fired the first shot of automatic driving

Fully automatic parking system, on-board camera and ultrasonic radar fusion.Fully automatic parking, usually use cost-effective ultrasonic sensors, and 360 surround view system. There are generally 12 ultrasonic sensors, 4 reversing radars, 4 parking assist ultrasonic radars and 4 parking assist ultrasonic radars. They emit ultrasonic signals and then receive signals reflected from obstacles. The length of time to evaluate the distance to the obstacle. Among them, the front and rear radars are used for ranging, and the left and right radars are used to detect the length and width of the parking space. Now more advanced automatic parking system, will be combined with the selection of millimeter wave radar system, distance detection and anti-interference ability is stronger. The 360 surround view system is used to identify the parking auxiliary line, and at the same time, it can also let the driver know the situation around the vehicle in the car, and can also personally intervene in the parking action if necessary. Fully automatic parking is a typical ADAS function that can be realized by the fusion of two sensors, an on-board camera and an ultrasonic radar.

Automatic parking took the lead in the popularity of luxury brands and joint venture brands, but the differentiation between brands is very large, independent began to spread.We have combed in detail the situation of all mainstream brands in car home equipped with automatic parking and seating configuration. From the brand point of view, luxury cars are represented by BBA, which is almost universal in the whole system. Joint venture brands are very differentiated and basically popular. However, Toyota, Mazda, Kia and Citroen and other four brands are not equipped in the whole system, and autonomy has just begun to be popularized. From the price point of view, the high-end models of their respective brands basically have automatic parking function, the top of the mid-end models have automatic parking function, and the low-end models are basically optional or not. In terms of joint venture and autonomy, Germany and the United States are the leaders in the joint venture to popularize the automatic parking function, while Japan, South Korea and the legal system are far behind. Geely, Great Wall and Chang 'an are the pioneers in popularizing the automatic parking function independently, while SAIC, GAC and the first three lines are far behind. Judging from the plans of various car companies, many of the new cars or replacement models launched this year will be equipped with automatic parking functions. At the same time, the market share of downstream car companies is also expected to be concentrated to advantageous manufacturers, and the automatic parking function is expected to accelerate in the future. Fully automatic parking is one of the basic functions of automatic driving, has been the first to begin to popularize, the future high-speed automatic driving is also expected to begin to gradually load the road.

7.3.AI computing and storage improvement bring systematic improvement opportunities

AI algorithm and CV chip help automatic driving

AI computing power has increased 300000 times in 6 years, and deep learning has boosted machine vision accuracy.Traditional machine vision mainly completes image classification and detection through search algorithms, edge algorithms, Blob analysis, caliper tools, optical character recognition and color analysis, and with the addition of AI algorithms, the application scenarios and recognition performance of machine vision have been greatly improved. At present, in AI graphics processing, the efficient large convolution deconstruction and multiplexing mechanism has matured, and terminal AI computing can further relieve bus bandwidth pressure and improve system efficiency. In terms of decision planning, decision tree, Bayesian network and other methods have long been widely used, in recent years, deep convolutional neural network and deep intensive learning enable AI to optimize centralized neural network through a large number of learning, with the traditional decision-making model perfect, can achieve complex conditions of decision-making.

AI algorithms and automotive CV chips make self-driving cars possible.Computing platform chips for vehicles should have the characteristics of reliable performance, low power consumption and strong image computing power. Although traditional GPU and FPGA chips have high versatility, they have high energy consumption. With the continuous development of semiconductor technology and the improvement of process performance, ASICS chips for the vehicle market emerge as the times require. Deep learning and neural network IP can be directly solidified in CV chips more efficiently, and specialized vehicle vision chips can save redundant structures, improve the calculation efficiency of unit energy consumption. The raw data received by the on-board sensor is through the image processing module, through the deep learning algorithm, with the special AI chip, to achieve high-precision environmental perception. Environmental data through the decision and planning network, in turn through the global decision planning, local trajectory planning and chassis execution control to achieve automatic driving path decision-making, and in the reinforcement learning to achieve behavior prediction and intelligent obstacle avoidance.

Storage performance improvement to meet the requirements of automatic driving data transmission

High-bandwidth DRAM improves ADAS system bus transfer capability.High-speed DDR granules achieve high bandwidth through data encoding and read-write separation, and high capacity by virtue of high-density addressing capability. The ADAS system in the L5 era requires high bandwidth of the data bus between the sensor and the main control unit, reaching 300 GB/s. High-speed DDR can break through the bandwidth bottleneck and can operate under high temperature and harsh conditions related to automobiles.

3D NAND solves the problem of large-capacity non-volatile storage in self-driving cars.Compared with traditional mechanical hard disks, 3D NAND Flash has the advantages of high storage density, high reading and writing speed, and high stability, which solves the problem that storage capacity and reliability are difficult to balance. It can not only meet the high reading and writing speed and storage space requirements of massive data in the era of automatic driving (up to 1 trillion byte by 2020), but also can work stably in more complex situations. As the number of vertical layers increases in the future, 3D NAND storage will continue to improve performance and storage capacity, further meeting the needs of autonomous vehicles.

NOR Flash meets the need for "immediacy" in self-driving cars.NOR Flash can execute program code on its own without DRAM, and the program startup speed is much higher than the NAND-DRAM combination. This instant start capability is suitable for systems that require start speed in autonomous driving, such as instrument cluster systems, infotainment systems and ADAD systems. At the same time, NOR flash memory can work under more extreme conditions than 3D NAND, and work stably under harsh conditions such as car engines and chassis.

8. Automotive electronic sensors related to the target.

Comprehensive analysis of the various types of sensors involved in the company, we recommend the relevant targets are as follows:

Electronic industry:Weir shares, Shunlu Electronics and Wentai Technology;

Computer Industry: Four-dimensional Tuxin, Zhonghai Da and Wanji Technology.

Automotive industry: Desai Siwei and Baolong Technology.