Posted 2025-01-04 00:00:00 +0000 UTC
"The accuracy is not allowed. We can't let single eyes carry the pot any more." In the face of the increasingly civilian cost and closer to the mass production of the car gauge lidar, Zhou Shengyan, founder of maxieye, found that it is urgent to attack the monocular vision with the highest accuracy in the world. There is no doubt that the key technology restricting the development lies in the perception level. However, the industry is still criticizing the visual perception distance error, the lack of three-dimensional space modeling ability, and the sample database is too complex. Therefore, Zhou Shengyan decided to lead the algorithm team to redefine monocular vision and cross-border challenge lidar. A full year ago, maximeye began to deploy in advance with the help of Solomon 1.0 plan. At that time, the company's first task was to accurately identify the target in the image two-dimensional world, and called out the slogan of "open interface and underlying data enabling industry". According to official data, at present, there are more than 10 passenger car factories, more than 20 tier 1, more than 20 scientific research institutions and colleges and universities have become members and partners of Salomon plan. After exchanging more real vehicle data and fully exploring potential customers, maximeye has now released Salomon version 2.0. At the same time, it has targeted optimization according to the failure of different urban scenes, special lane lines and strong light interference and other issues mentioned in customer feedback. The latest products will continue to open protocol architecture, image data, control strategy, and adhere to the strategy of customer customization. At the same time that Salomon 2.0 was released, the product was also open to customers for application, and the car factory could directly test it. In terms of product details, Zhou Shengyan revealed that the new products are suitable for passenger cars, buses and heavy trucks, so as to achieve full coverage of high, medium and low models. Support on-line automatic calibration. Through dynamic calibration compensation, the vehicle manufacturer can be competent for complex working conditions and reduce the possibility of misjudgment and failure without relying on calibration equipment during docking technology development. Cheyunxiong noted that the new generation of products uses the same 57 ° HD camera as the previous generation, providing 2 million pixel CMOS. That is to say, under the premise of the same cost, what the team completed during this period is to make the product have the ability of beyond visual range through algorithm design. In general, maxieye is based on the three-dimensional construction of precise ranging to build Monocular Stereo vision. The new product can achieve point cloud scanning close to radar accuracy within 50m, and can provide the direct ranging function of the target. As for the rule target that customers are most concerned about, it supports vehicle detection within 200m, pedestrian and small target detection within 100m, and provides accurate three-dimensional attitude tracking function. On the whole, under the support of this technology, the error between recognition effect and lidar is less than 1%. Compared with the benchmark level technology infrastructure in the industry, maxieye hopes to raise the dimension from the perspective of depth estimation and slope estimation. Among them, the former is used for precise target location and 3D static scene construction, and the latter is less than 0.1 degree compared with lidar. In addition, due to the greatly improved detection accuracy, in the application level, the team also tries to solve the problems that most players cannot solve, such as ghost probe. What's more, with the opportunity of commercial vehicle mass production, this young enterprise has established its own data barrier. Maxieye holds the test data of 2 million kilometers from the real feedback of the main engine factory, which basically covers the comprehensive data of 700000 kilometers of the existing road conditions and the crash of 42 mass production models. It also rents test buses and runs 100000 kilometers of test mileage. By installing a camera every 10cm on the glass, the team gets the data of different angles from the perspective of the truck, and solves the corner case as much as possible. The change of chassis between full load and no load, more complex working conditions, all put forward higher requirements for product grinding. This makes the ranging error of the system keep in the range of 1% - 3%, which is far lower than the 6% required by the car factory. In addition to the basic ADAS functions, the car manufacturers also hope to offset the cost increase brought by the policy and regulations with the added value of some functions. Therefore, the company has developed several more practical functions, such as intelligent headlamp application, rain recognition to control the opening speed of wiper, recognition of slope to compensate the target distance, so as to actively control fuel saving of heavy truck diesel vehicles. Interestingly, maxieye's goal is not to kill the lidar. "The more cross-border a single sensor, the better vehicle safety redundancy." Zhou Shengyan explained to cheyunzhu. In this way, the monocular technology route advocated by the company has completed another freedom beyond the landing model - the car factory can independently choose more integrated products. More choice of millimeter wave radar space means that customers do not have to choose a higher cost supply scheme to compensate for visual performance. This is undoubtedly an excellent opportunity for the domestic millimeter wave radar manufacturers who miss the policy dividend in 2019 to enable the industry and activate the ecology. It should be noted that in the eyes of domestic ADAS manufacturers, 2020 is the first year of mass production of heavy truck AEB. In order to meet the year of realization, maxieye needs to make stable production and revenue, and conduct sufficient technical verification with the commercial vehicle actuator company before the formal production, so as to ensure that the market share of AEB in the heavy truck market next year can duplicate its achievements in the field of passenger cars, reaching more than 80%. At this stage, the company's products have been favored by 21 front loading mass production customers and 5 designated truck enterprises. Zhou Shengyan also disassembled the market opportunities at this stage in detail. The development trend of several national policies shows that AEB functions are developing from "recommended installation" standard to "forced installation" standard. Europe will force 100% of its models to be equipped with AEB in 2022, while the domestic time will not exceed 2025. Obviously, the vehicle installation standard is gradually transitioning from FCW early warning function to AEB and LKA in 2021. The main body of the market is also penetrating from buses to buses. The next step is the truck field with a market scale of more than 3 million, including 800000-900000 heavy trucks, light trucks and light medium trucks, and finally covers the passenger vehicle field with a capacity of 20 million. According to the plan, maximeye will release 30000 products in 2019. In order to meet the new round of market competition, the company also expanded the capacity of the factory in Chuansha, with the ultimate capacity of 1.3 million units, 1-2 times higher than the expected design.. Zhou Shengyan is quite optimistic about the next step of development. In the face of the statement of "eating policy dividend", he would not deny: "policy is fair to everyone." The company is precisely to grasp this time window and carry out technical reserve through policies.
Copyright © 2020. TUTESL All rights reserved.