Autonomous driving is no longer a piece of paper talk. Through the efficient and stable cooperation of dozens of parts and components at the three levels of perception, decision-making and execution, autonomous vehicles have gradually realized the landing application of different scenarios and different levels.
At present, there are two main routes for the automatic driving perception layer program: one is the multi-sensor fusion scheme, which is widely recognized and promoted by many OEMs; the other is the vision-based technical route, interestingly, based on the safety and reliability of automatic driving, there is also a controversy between two technical directions, namely the pure visual perception represented by Tesla, and the lidar perception favored by new car companies.
The technical routes are not the same, but with the rapid development of automatic driving, on-board sensors "rise by the wind" to achieve product acceleration and innovation upgrades. While optical manufacturers grasp the needs of upstream car companies, they also need to ensure the reliability, safety and durability of on-board sensors through technological innovation and research and development strength. Chen Bo, head of of OFILM's intelligent vehicle algorithm, attended the second intelligent vehicle visual perception summit and discussed the advanced road of domestic vision suppliers of autonomous driving with a number of enterprises upstream and downstream of the industrial chain.
Solve the problem of the working environment of the car camera
In terms of environmental perception, the sensors for the main applications of autonomous driving include cameras, millimeter-wave radar, lidar, ultrasonic radar, etc. At the same time, car sensors can also be classified into different perception application scenarios such as driving assistance, parking assistance and in-vehicle personnel monitoring. In general, the number of sensors equipped with a car directly determines the level of intelligence.
As the core sensor of ADAS, the on-board camera can recognize a variety of characteristics of objects, and realize functions such as lane departure warning (LDW), distance detection (HMW), and traffic sign recognition (TSR). Throughout the whole process of vehicle driving to parking, automatic driving has higher requirements for the characteristics of high temperature resistance, anti-magnetic, earthquake resistance and long service life of the on-board camera. Chen Bo said that with the increase in the number of car camera installations, due to the limited power supply of batteries and the limited space for installing cameras, the demand for vehicle cameras to further reduce the size of the circuit board and reduce power consumption is increasing. The hard "arms race" of on-board cameras has begun.
Chen Bo mentioned that in order to solve the heat dissipation problem of the vehicle camera module, OFILM proposed four thermal management solutions such as vacuum cavity vapor plate heat dissipation technology, laser welding, and temperature drift compensation.
First of all, OFILM improves the packaging technology, using COB packaging technology, directly compressing the photosensitive chip to the PCB board, compared with the traditional CSP packaging technology, greatly improving the heat dissipation effect. OFILM COB technology has been maturely applied in the field of mobile phones, and will "translate" the technical advantages of mobile phone business to the field of vehicle business.
Secondly, OFILM introduced the vacuum chamber heat dissipation plate heat dissipation technology, broke through the traditional laying copper mesh or copper powder sintered capillary core structure, and adopted the precision etching microstructure integrated capillary core process, successfully developed a 0.3mm ultra-thin vacuum chamber vapor heat plate, and reached the stable process level that can be mass-produced.
At the same time, OFILM abandons the traditional screw fixing method and adopts laser welding, which greatly reduces the number of related components, thereby improving the reliability and consistency of the camera.
Finally, the company has also developed a set of temperature compensation schemes and internal reference calibration (the distortion of the camera is calibrated through the algorithm software to make the imaging effect clearer and more realistic), and the internal reference calibration process to ensure the clarity and authenticity of the camera imaging effect to the greatest extent. Chen Bo mentioned for example that by selecting the optimized structure of the material and the effect of "thermal expansion and contraction", OFILM always maintains the camera within a certain temperature range, so as to achieve a good imaging effect.
According to Chen Bo, OFILM's 1 million pixel and 2 million pixel vehicle cameras have achieved mass production, and the 8 million pixel ADAS front-view hd camera is also about to be mass-produced. In addition, OFILM is also developing a front-view tri-eye camera assembly based on an 8-megapixel high-definition camera, which will meet the visual perception of "far, medium and near" three distances.
Millimeter-wave radar provides greater safety redundancy for autonomous driving
Some people in the industry believe that there are still certain shortcomings in pure visual perception. After Tesla launched a pure visual version of the automatic driving system FSD Beta V9.0, the voice of industry doubts is more intense, in the rain, snow, smog and other harsh conditions, the automatic driving function based on vision technology will be greatly limited, abandon the cost is not high but can solve the interference problem of millimeter wave radar, is considered to be a radical choice that is not worth the loss. Millimeter-wave radar, with its advantages of accurate ranging and all-weather work, is regarded as a "means of making up for the shortcomings of visual technology". Using the integration of vision technology and millimeter wave radar technology, the landing of automatic driving technology is more "economical and practical", and it can also meet the safety needs of L2 and L3 levels of automatic driving.
According to Chen Bo, in addition to the vehicle camera, OFILM also laid out other sensor technologies such as millimeter-wave radar, and provided vision-based deep learning algorithms, multi-sensor fusion SLAM algorithms, millimeter-wave radar algorithms, and so on. In the field of millimeter-wave radar, OFILM has released a new generation of short-range millimeter-wave radar (77GHz), and based on this millimeter-wave radar, it has focused on the introduction of solutions for various scenarios such as reversing assist system (PAS), cabin live detection system, automatic parking (APA)/valet parking (AVP) and so on.
Next, OFILM will also launch medium and long-range millimeter wave radar and 4D imaging radar to meet the sensor needs of higher-level autonomous driving systems.
For the low-speed solution of millimeter-wave radar, Chen Bo said that there are five major advantages. First, there is a more accurate obstacle positioning; second, there is a greater range of perception; third, it has a higher driving speed; fourth, static obstacles and dynamic obstacles are distinguished; fifth, there are richer positioning means.
At the same time, Chen Bo also said that OFILM's research and development of "not the same as the market" lidar, will use a pure solid-state solution, there is no internal rotation structure, can maximize the safety and reliability of lidar.
Great perceptual fusion is still the mainstream direction
Some insiders have said that it is difficult to achieve automatic driving simply through visual technology, "After so many years of exploration, the technical solution of large perception integration has become the general consensus of the industry." "Data fusion from multiple sensors can minimize the drawbacks of a single sensor, improving the reliability and robustness of the system through system redundancy.
As a supplier of full-stack intelligent driving solutions, OFILM has made great efforts at the sensor level, and has also made progress in the field of controllers and algorithms. In terms of intelligent driving domain controllers, OFILM will launch a domain controller solution with a computing power of more than 140T, which can support 16 channels of CAN, has 12 high-speed Ethernet interfaces, and can access 13 cameras.
At this stage, ofimonic algorithm team has developed a perceptual fusion algorithm that meets L3+ level automatic driving, and is committed to the research and development of new technologies, integrating weak supervision learning algorithms into automatic driving products, and solving the problem that traditional strong supervision deep learning algorithms must rely on a large number of manual labeling work.
Overall, from next-generation sensors to various system solutions, OFILM will become a full-stack solution provider of intelligent driving systems with cameras, ultrasonic sensors, millimeter wave radar, lidar, 4D imaging radar, intelligent driving domain controllers, system software and algorithms.
According to public information, OFILM has begun to lay out intelligent vehicles since 2015, and has obtained the qualification of first-class suppliers of more than 20 domestic and foreign car companies. According to of THEILM's semi-annual report, the company is based on advanced sensors, but not limited to the perception area of on-board cameras, millimeter wave radar, and lidar, and the company will continue to actively expand hardware, software and algorithms, and the company has the ability to provide automatic driving, intelligent central control, and body control system level solutions. In the first half of 2021, the company's intelligent vehicle business grew significantly, achieving operating income of 413 million yuan, an increase of 78.22% year-on-year; comprehensive gross profit margin of 14.83%, an increase of 0.28 percentage points year-on-year. Among them, ADAS (Advanced Driver Assistance System) achieved operating income of 120 million yuan, an increase of 777.68% year-on-year. Pacific Securities Research Report once mentioned that "OFILM through epitaxial expansion and epitaxial mergers and acquisitions, to create a new engine of "two-wheel drive", focusing on the development of HMI, ADAS, ECU and other automotive electronics business, in accordance with the path from hardware to software, from product manufacturing to content services gradually layout the field of intelligent vehicles." ”
For optical manufacturers, the new market for smart cars is creating huge business opportunities. However, optical lenses are a high-barrier industry, and the barriers to vehicle-grade products are even higher. OFILM and other optical leading enterprises have opened a new chapter in the field of intelligent vehicles, and if you want to be an excellent automatic driving supplier, the head company still needs to continuously increase research and development efforts in order to occupy the main part of the value chain with technical advantages and stand out in the new track.
This article does not constitute any investment advice, investors act accordingly at their own risk.