Zosi Automobile Research Institute released the "2019-2020 Automobile Vision Industry Chain Research Report (Part 2) Binocular and Others".
According to Sunny's optical point of view, automotive cameras are divided into perception cameras and imaging cameras.
Perception cameras are used for active security and require accurate capture of images, which are generally used for forward and intravision. Image cameras are used for passive security and store or send the captured images to users, generally used for surround view and rear view. Therefore, the perception camera and the image camera are completely different in terms of imaging quality requirements and temperature reliability requirements.
perception camera is used for lane detection, signal light detection, road sign recognition, in-vehicle monitoring, etc. If any errors in the image taken by the perception camera will cause software calculation errors and lead to inevitable consequences. Therefore, the price sensitivity of perceptual cameras is relatively low.
image cameras are very sensitive to prices, so there is a fierce price war, and the performance requirements are not very high. Domestic companies have more interventions, but most of them are not profitable. From the perspective of the market structure of China's passenger car surrounding front-mounted market in 2019, it can be seen that the market share of various manufacturers in the surrounding market is relatively scattered, unlike the front-view monocular market, which shows that the market is relatively concentrated, the TOP6 accounts for more than 90%.
Source: Zosi Auto
Front view camera requires complex algorithms and chips. The unit price for a single front view is close to 1,000 yuan, and for a double eye, it costs more than 1,000 yuan. The unit price of rear view, side view and built-in camera is around 200 yuan.
Breakthrough in the binocular field
Domestic start-ups have made breakthrough progress in the binocular field, especially in the commercial vehicle and special vehicle markets.
Zhongke Huiyan Binocular has shipped more than 10,000 units, mainly used in Apollo series unmanned vehicles, Jiangling Light Vehicles, sanitation vehicles, unmanned ships, tractors, patrol vehicles, etc. Typical applications of Zhongke Eye Binocular Products are AEBS and height-limit detection. Bus groups in about 30 cities have assembled the Zhongke Eye Binocular System.
binoculars are used for the detection of height limit devices, and are the first of the first of Zhongke Huiyan. In addition to conventional height limit poles, various non-standard height limit devices such as small archways in the countryside, culverts, etc. can accurately sense and promptly remind drivers through sound and light alarm devices. RVs, special vehicles, etc. all have strong demand for limited height testing.
Source: Zhongke Huiyan
In 2019, Hammerhead Shark announced the development of a free binocular system, which can use two completely independent monocular cameras to realize binocular systems to get rid of the inherent defects such as large size, complex process, difficult installation and high cost of binocular equipment.
Source: Hammerhead Shark Technology
Two cameras are independently installed and fixed on the right side of the car body. There is no rigid connection in the middle, nor does it require strict control of the angle and spacing of the camera. The core principle of "free binocular" is self-calibration technology. Even if the camera undergoes slight deformation and movement during use, the Hammerhead's algorithm will automatically detect and recalibrate, eliminating the regular recalibration and calibration required for conventional binoculars.
In April 2020, Foresight announced a partnership with FLIR to integrate FLIR's infrared cameras, combining visible and thermal stereo vision technologies to provide accurate obstacle detection under harsh light and weather conditions. The data fusion between the two three-dimensional channels can effectively solve the problem of non-report or false alarms in extreme cases such as tunnel entrances where light and dark changes are rapidly changed.
Zoth's millimeter wave report pointed out that millimeter wave radar is eroding the territory of other sensors. The same is true for cameras. Many companies use monocular cameras to do ranging and do 3D imaging, trying to replace binocular or lidar . For example, MAXIEYE.
MAXIEYE The first generation product IFVS-200 series is based on machine learning solutions, and the third generation product IFVS-500 series is based on deep learning solutions to achieve monocular ranging and 3D scanning. The IFVS-500 series allows monocular vision products to perform three-dimensional scanning like lidar, or to achieve 3D scene point cloud scanning close to lidar within 50 meters, providing direct ranging function of targets, realizing motor vehicle detection in a range of 200 meters, pedestrian and small target obstacle detection in a range of 100 meters.
has enhanced the visual ability in extreme scenarios
Whether inside or outside the car, visual ability needs to be obtained when the light is poor or even dark, which means that infrared technology is needed. ON Semiconductor's RGB-IR image sensor uses NIR (near infrared) technology, while another manufacturer, Trieye, uses a short-wave infrared (SWIR) camera. The advantage of short-wave infrared cameras is that they can see objects in any weather/light conditions, and can identify road hazards in advance (such as road icing).
In April 2020, Alibaba Damo Academy developed an ISP processor for vehicle camera . According to the road test results, using the ISP processor of the DAMO Academy, the image object detection and recognition capability of the vehicle camera in the night scene is significantly improved by more than 10% compared with the mainstream processors in the industry, and the originally vague markers can also be clearly identified.
On May 19, 2020, Howie Technology officially released the image sensor OX03A2S equipped with Nyxel near-infrared technology. This 2.5-megapixel ASIL-B grade sensor is designed for external imaging applications and can be used in low-light or even light-free environments within 2 meters of the body. The OX03A2S can detect and identify objects that other image sensors cannot capture in low-light environments.
Source: Howe Technology
Visual perception enters the deep water zone, and the algorithm determines the winner
With the large increase in vision sensor in smart cars, the addition of different types of vision sensors has generated huge data, which brings challenges and opportunities to algorithm processing.
Visual ADAS leader in the visual perception system Mobileye In the visual perception system, a variety of independent perception algorithms are used to achieve redundant superposition, with the purpose of improving the accuracy and stability of perception in the two dimensions of Detection and Measurement. Detection determines what object the perceived object is. Measurement calculates the 2D picture of the camera to obtain 3D information of the perceived object.
In the Detection dimension, Mobileye uses 6 independent algorithms:
- 3D Auto Detection (3DVD): recognizes the target vehicle in the 2D screen and marks it on the 3D bounding box.
- Full Image Detection: It is mainly used to identify large objects (such as passenger cars or trucks) at close distances on both sides of the vehicle.
- Top View FS: Focus on identifying the roads that are not occupied in the screen and marking them.
- Features Detection (such as Wheels): Focus on identifying objects with unique features, such as wheels.
- VIDAR: Generate a 3D picture through triangulation of multiple cameras, and then import the 3D picture into the lidar perception algorithm for object recognition.
- Scene Segmentation (NSS): Through pixel-level recognition, different types of objects are divided and marked with different colors.
Source: Mobileye
Tesla is also a leader in visual perception algorithms. Tesla calls its deep learning network HydraNet. The entire HydraNet contains 48 different neural networks. Through these 48 neural networks, 1,000 different prediction vectors can be output. Theoretically, HydraNet can detect 1,000 objects at the same time. To enhance algorithm capabilities, Tesla specially acquired computer vision startup DeepScale.
In order to catch up with Tesla and Mobileye in visual perception algorithms, OEMs and Tier1 are expanding their team of software engineers. Algorithm capability will become one of the decisive factors that influence the quality of visual perception performance.
Zosi Automobile Research Institute released the "2019-2020 Automobile Vision Industry Chain Research Report (Part 2) Binocular and Others".
According to Sunny's optical point of view, automotive cameras are divided into perception cameras and imaging cameras.
Perception cameras are used for active security and require accurate capture of images, which are generally used for forward and intravision. Image cameras are used for passive security and store or send the captured images to users, generally used for surround view and rear view. Therefore, the perception camera and the image camera are completely different in terms of imaging quality requirements and temperature reliability requirements.
perception camera is used for lane detection, signal light detection, road sign recognition, in-vehicle monitoring, etc. If any errors in the image taken by the perception camera will cause software calculation errors and lead to inevitable consequences. Therefore, the price sensitivity of perceptual cameras is relatively low.
image cameras are very sensitive to prices, so there is a fierce price war, and the performance requirements are not very high. Domestic companies have more interventions, but most of them are not profitable. From the perspective of the market structure of China's passenger car surrounding front-mounted market in 2019, it can be seen that the market share of various manufacturers in the surrounding market is relatively scattered, unlike the front-view monocular market, which shows that the market is relatively concentrated, the TOP6 accounts for more than 90%.
Source: Zosi Auto
Front view camera requires complex algorithms and chips. The unit price for a single front view is close to 1,000 yuan, and for a double eye, it costs more than 1,000 yuan. The unit price of rear view, side view and built-in camera is around 200 yuan.
Breakthrough in the binocular field
Domestic start-ups have made breakthrough progress in the binocular field, especially in the commercial vehicle and special vehicle markets.
Zhongke Huiyan Binocular has shipped more than 10,000 units, mainly used in Apollo series unmanned vehicles, Jiangling Light Vehicles, sanitation vehicles, unmanned ships, tractors, patrol vehicles, etc. Typical applications of Zhongke Eye Binocular Products are AEBS and height-limit detection. Bus groups in about 30 cities have assembled the Zhongke Eye Binocular System.
binoculars are used for the detection of height limit devices, and are the first of the first of Zhongke Huiyan. In addition to conventional height limit poles, various non-standard height limit devices such as small archways in the countryside, culverts, etc. can accurately sense and promptly remind drivers through sound and light alarm devices. RVs, special vehicles, etc. all have strong demand for limited height testing.
Source: Zhongke Huiyan
In 2019, Hammerhead Shark announced the development of a free binocular system, which can use two completely independent monocular cameras to realize binocular systems to get rid of the inherent defects such as large size, complex process, difficult installation and high cost of binocular equipment.
Source: Hammerhead Shark Technology
Two cameras are independently installed and fixed on the right side of the car body. There is no rigid connection in the middle, nor does it require strict control of the angle and spacing of the camera. The core principle of "free binocular" is self-calibration technology. Even if the camera undergoes slight deformation and movement during use, the Hammerhead's algorithm will automatically detect and recalibrate, eliminating the regular recalibration and calibration required for conventional binoculars.
In April 2020, Foresight announced a partnership with FLIR to integrate FLIR's infrared cameras, combining visible and thermal stereo vision technologies to provide accurate obstacle detection under harsh light and weather conditions. The data fusion between the two three-dimensional channels can effectively solve the problem of non-report or false alarms in extreme cases such as tunnel entrances where light and dark changes are rapidly changed.
Zoth's millimeter wave report pointed out that millimeter wave radar is eroding the territory of other sensors. The same is true for cameras. Many companies use monocular cameras to do ranging and do 3D imaging, trying to replace binocular or lidar . For example, MAXIEYE.
MAXIEYE The first generation product IFVS-200 series is based on machine learning solutions, and the third generation product IFVS-500 series is based on deep learning solutions to achieve monocular ranging and 3D scanning. The IFVS-500 series allows monocular vision products to perform three-dimensional scanning like lidar, or to achieve 3D scene point cloud scanning close to lidar within 50 meters, providing direct ranging function of targets, realizing motor vehicle detection in a range of 200 meters, pedestrian and small target obstacle detection in a range of 100 meters.
has enhanced the visual ability in extreme scenarios
Whether inside or outside the car, visual ability needs to be obtained when the light is poor or even dark, which means that infrared technology is needed. ON Semiconductor's RGB-IR image sensor uses NIR (near infrared) technology, while another manufacturer, Trieye, uses a short-wave infrared (SWIR) camera. The advantage of short-wave infrared cameras is that they can see objects in any weather/light conditions, and can identify road hazards in advance (such as road icing).
In April 2020, Alibaba Damo Academy developed an ISP processor for vehicle camera . According to the road test results, using the ISP processor of the DAMO Academy, the image object detection and recognition capability of the vehicle camera in the night scene is significantly improved by more than 10% compared with the mainstream processors in the industry, and the originally vague markers can also be clearly identified.
On May 19, 2020, Howie Technology officially released the image sensor OX03A2S equipped with Nyxel near-infrared technology. This 2.5-megapixel ASIL-B grade sensor is designed for external imaging applications and can be used in low-light or even light-free environments within 2 meters of the body. The OX03A2S can detect and identify objects that other image sensors cannot capture in low-light environments.
Source: Howe Technology
Visual perception enters the deep water zone, and the algorithm determines the winner
With the large increase in vision sensor in smart cars, the addition of different types of vision sensors has generated huge data, which brings challenges and opportunities to algorithm processing.
Visual ADAS leader in the visual perception system Mobileye In the visual perception system, a variety of independent perception algorithms are used to achieve redundant superposition, with the purpose of improving the accuracy and stability of perception in the two dimensions of Detection and Measurement. Detection determines what object the perceived object is. Measurement calculates the 2D picture of the camera to obtain 3D information of the perceived object.
In the Detection dimension, Mobileye uses 6 independent algorithms:
- 3D Auto Detection (3DVD): recognizes the target vehicle in the 2D screen and marks it on the 3D bounding box.
- Full Image Detection: It is mainly used to identify large objects (such as passenger cars or trucks) at close distances on both sides of the vehicle.
- Top View FS: Focus on identifying the roads that are not occupied in the screen and marking them.
- Features Detection (such as Wheels): Focus on identifying objects with unique features, such as wheels.
- VIDAR: Generate a 3D picture through triangulation of multiple cameras, and then import the 3D picture into the lidar perception algorithm for object recognition.
- Scene Segmentation (NSS): Through pixel-level recognition, different types of objects are divided and marked with different colors.
Source: Mobileye
Tesla is also a leader in visual perception algorithms. Tesla calls its deep learning network HydraNet. The entire HydraNet contains 48 different neural networks. Through these 48 neural networks, 1,000 different prediction vectors can be output. Theoretically, HydraNet can detect 1,000 objects at the same time. To enhance algorithm capabilities, Tesla specially acquired computer vision startup DeepScale.
In order to catch up with Tesla and Mobileye in visual perception algorithms, OEMs and Tier1 are expanding their team of software engineers. Algorithm capability will become one of the decisive factors that influence the quality of visual perception performance.
《2019-2020 Automotive Vision Industry Chain Research Report (Part 2) Binocular and Others》Catalogue
05
Domestic Automotive Vision Enterprise Research
5.1 Zhongke Huiyan
5.1.1 Zhongke Huiyan Introduction
5.1.2 Zhongke Huiyan products
5.1.3 Zhongke Huiyan development
5.1.4 Comparison of similar products
5.1.5 Production and manufacturing
5.1.6 Binocular stereoscopic visual product application
5.2 Hammerhead Shark Technology
5.2.1 Introduction to Hammerhead Shark Technology
5.2.2 Hammerhead Shark Product
5.2.3 Hammerhead Shark Free Binocular System
5.2.4 Big Truck Assisted Safety Driving System
5.3 Xiaomi Intelligent
5.3.1 Introduction to Xiaomi Intelligent
5.3.2 Xiaomi Intelligent Core Technology
5.3.3 Xiaomi Bionic Camera Depth Series
5.3.4 Xiaomi Intelligent Binocular ADAS Solution
5.3.5 Xiaomi intelligent unmanned delivery solution
5.4 Yuansu Technology
5.4.1 Introduction to Yuansu Technology
5.4.2 Yuansu Technology Solution
5.4.3 Yuansu Technology Product Application Areas
5.5 Huaxin Technology
5.5.1 Introduction to Shenzhen Huaxin Technology Co., Ltd.
5.5.2 Huaxin Technology Tech Tech Tech
5.5.3 Huaxin Technology Tech Tech Tech Tech Tech Tech
5.5.4 Huaxin Technology Partner
5.6 Hisense Network Technology
5.6.1 Introduction to Qingdao Hisense Network Technology
5.6.2 Hisense Network Technology Solution
5.6.3 Hisense Network Technology Binocular ADAS
5.7 Dahua Technology
5.7.1 Introduction to Zhejiang Dahua Technology
5.7.2 Dahua Shares Products
5.7.3 Dahua Biological Application
5.7.4 Latest News
5.8 Foresight 5
5.8.1 Foresight Introduction
5.8.2 Foresight quadrature vision system
5.8.3 Binocular vision system Eyes-On
5.8.4 Foresight Eye-Net
5.8.5 Foresight development dynamics
5.9 Japan Ricoh
5.9.1 Japanese Ricoh Introduction
5.9.2 Japanese Ricoh Development Strategy
5.9.3 Japanese Ricoh Binocular Module
5.9.4 Other Applications of Japanese Ricoh Binocular Products
5.10 Other Tier1's binocular products
5.10.1 Mainland Automobile
5.10.2 Bosch binocular camera
5.10.3 Electric stereo vision sensor
5.10.4 ZF binocular camera
5.111 Binocular product technology summary
5.11.1 Technical development trend of binocular vision
5.11.2 Binocular chip
06
Other automotive vision companies research
6.1 Overview of the surrounding view market
6.1.1 Number of installed cameras in China in 2018-2019
6.1.2 Number of installed cameras in China in 2018-2019
6.1.3 The proportion of assembly volume of new domestic cars in China in 2019 and the TOP20 brand
6.1.4 The assembly rate of new domestic cars in China in 2019
6.1.5 2019 2019 surround view cameras in China in 2019 surround view cameras in China in 2019
6.2.1 Introduction to Tongzhi Electronics
6.2.2 Global layout and product layout
6.2.3 Camera products
6.2.4 Main products and main customers
6.3 Hangsheng
6.3.1 Introduction to Hangsheng Electronics
6.3.2 Hangsheng product system
6.3.3 Hangsheng external cooperation
6.3.4 Hangsheng supplier
6.3.5 Hangsheng customers
6.4 Hikvision
6.4.1 Introduction to Hikvision
6.4.2 Revenue of Hikvision
6.4.3 Automotive electronic products
6.4.4 Hikvision visual products
6.5 Shengtek
6.5.1 Introduction to Shengtek
6.5.2 Shengtek Automotive Surround View System
6.5.3 Shengtek Main Camera Products
6.6 Otermin
6.6.1 Company Profile
6.6.2 ADAS Products
6.6.3 Otermin core technology
6.6.4 camera product
6.6.5 automatic parking system
6.6.6 Customers and partners
6.7 Minshi Group
6.7.1 Company profile
6.7.2 revenue
6.7.3 R&D and design capabilities
6.7.4 Camera business development history
6.7.5 Surrounding product
6.7.6 ADAS product application
6.8 LG
6.8.1 Company profile
6.8.2 LG automotive electronic product line
6.8.3 LG camera product
6.8.4 LG ADAS product
6.9 Panasonic
6.9.1 Company profile
6.9.2 Panasonic automotive business revenue
6.9.3 Automotive electronic product series
6.9.4 Automotive visual product
6.10 Valeo
6.10.1 Valeo Introduction
6.10.2 Valeo's revenue status in 2019
6.10.3 Valeo's automotive vision and ADAS system
6.10.4 Valeo's visual products
6.10.5 Valeo's visual product partners
6.11 Other
07
Enterprise research on key components of automotive cameras
7.1 ON Semi
7.1.1 Introduction to ON Semi
7.1.2 2019 ON Semi business status and automobile business layout
7.1.3 ON Semi-Automobile ecosystem
7.1.4 ON Semi-Automobile CMOS market position
7.1.5 ON Semi-AdAS image sensor
7.1.6 ON Semi-Automobile camera system architecture and CMOS product system
7.1.7 ON Semi-Automobile CMOS product
7.1.8 ON Semi-Automobile industry acquisition
7.2 Sony
7.2.1 Sony introduction
7.2.2 2019 Sony business status
7.2.3 Sony's technical advantages in the CMOS field
7.2.4 Sony camera module product system
7.2.5 Sony's automotive CMOS development history
7.2.6 Sony's automotive CMOS product line
7.2.7 Sony's automotive CMOS product
7.2.8 Sony's latest visual CMOS product
7.3 Howe Technology
7.3.1 Introduction to Howe Technology
7.3.2 Howe Image Sensor
7.3.3 Howe Automotive Image Sensor
7.3.4 The latest progress of Howe Technology
7.4 Original Phase Technology
7.4.1 Introduction to Original Phase Technology
7.4.2 CIS and chip products and planned development products
7.4.3 PixArt product
7.4.4 Original Phase Automotive Vision Application System
7.4.5 Automotive Gesture Control IC
7.5 Sunny Optics
7.5.1 Introduction to Sunny Optics
7.5.2 Sunny Optics revenue in 2019
7.5.3 Sunny Optics business layout
7.5.4 Sunny Automotive Lens Development History
7.5.5 Sunny's automotive lens sales and shipments
7.5.6 The latest trends and layout of Sunny Optics
7.6 Lianchuang electronic
7.6.1 Introduction to Lianchuang Electronics
7.6.2 Lianchuang Electronics' development history and shareholding structure
7.6.3 Lianchuang Electronics' operating income and R&D investment
7.6.4 Lianchuang Electronics' vehicle lens product
7.6.5 Lianchuang Electronics' application in the field of automotive camera
7.6.6 Lianchuang Electronics' main customer
08
development trend of the automotive vision industry
8.1 ToF application in automobiles
8.1.1 What is 3D perception technology
8.1.2 3D perception technology: stereo vision and structured light
8.1.3 3D perception technology: ToF
8.1.4 ToF potential
8.1.5 Main ToF image sensor comparison
8.1.6 Infineon and Panasonic's ToF
8.1.7 ADI ToF
8.1.8 Panasonic's new ToF image sensor
8.2 AI application in automotive vision
8.3 Other automotive vision development trends
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-