Just three years after its founding, Maxieye, a domestic automaker, made a bold decision to announce plans for Solomon
Behind the choice of openness, Maxieye CEO Zhou Shengyin told Leifeng's Xinzhijia that it was not a rash decision. The choice of openness meant that Maxieye had enough confidence in his products and really wanted to change the status quo of the industry in which customers could not get the underlying data.
In the silence, Zhou Shengyin and Leifeng Xinzhijia talked about the original intention of opening up, how to open up and the business path after opening up.
Comparison with industry benchmarking
Benchmarking with industry benchmarking is one of the common inspection standards in intelligent driving industry, but the methods of benchmarking are often different, so it does not have complete significance.
At present, one of the industry's standard-pointing methods is through application-based function indicators, such as comparing TTC,THW and LDW, MaxieyeCEO Zhou Shengyan told Lei Feng net Xinzhi Jia: "these indicators are reluctantly used in support of alarm applications with relatively low technical barriers. In fact, behind these indicators a lot of computational power is used to guarantee the function of simple index.
Zhou Shengyan told Lei Feng net new smart driver that at present, some technology companies in the industry claim that their automatic driving system can identify 1000 meters of vehicles and pedestrians, but they do not emphasize the lens parameters or CMOS resolution, so this comparison does not have the meaning of contrast. In addition, benchmarking comparisons on different computing platforms are also unscientific.
Maxieye has set up a fair benchmarking platform in order to more clearly identify its own product standards. According to Zhou Shengyin, Maxieye's products use CMOS (1280*720) with the same resolution of industry benchmark, similar lens HFOV52, and ASIC chip with comparable calculation power.
Maxieye shared the benchmarking results with Xinzhi Driver of Leifeng Networks (Public No. Leifeng Networks). The test was carried out on city roads and elevated roads with clear daytime and night. The number of lanes detected by benchmarking enterprises was 4; the number of lanes detected by ADVS-400 was 8; and the detection distance of lanes by benchmarking enterprises was 80. M; ADVS-400 is 100 m; 50 m lateral error and front wheel lateral error are basically the same.
In addition, Maxieye also makes the differential function processing for the special road condition in China, such as 3D recognition ability, occlusion target recognition, small target recognition. Zhou Shengyan also introduced the vision scheme of Maxieye to Lei Feng net Xinji.
Zhou Shengyan further explained to Leifeng New Intelligent Driver: TI TDA2X is actually a SOC architecture. In the near future, GPU and FPGA solutions will gradually marginalize the application in vehicle. The mainstream solution is NPU and general CPU designed by different architecture platforms, which can achieve very ideal results. For example, TI's TDA2X can use very small calculation power. In the environment of high density people on the Bund, the identification of highway urban roads, bends, road edges and car lights.
Open road choice
After polishing a range of product skills, Maxieye made a bold decision
Maxieye Yang Tengfei said to Xinzhijia: One of the pain points of this industry is that customers can only get very high-level data, application-level data, but the visual results of the bottom perception are difficult to get, and it is very difficult to do secondary development. Therefore, Maxieye hopes to break this dilemma.
According to Maxieye, the content of this open plan includes four aspects: open image data, tools and protocol architecture. Users include automatic driving customers, sensor customers, domain controller customers, actuator customers and low-speed vehicle customers. In addition, Yang added to Xinzhijia: Maxieye said that differentiated customization can be done for partners. He added some examples, such as identifying different lights (street lights and taillights) and judging the intentions of the vehicle during the driving process. At present, Maxieye has accumulated more than 20 customers, including passenger cars, commercial vehicles, Tie1 and other fields, but the specific names of customers are temporarily not easy to disclose.
Because users in different fields have different levels of vision requirements, Maxieye mainly extracts three general requirements: first, the rules and requirements of image formation; second, the application development in visual field coordinate system; third, the requirements of original image data.
Aiming at these main demands, Maxieye makes a point-to-point breakthrough, opens the output of the CAN bus interface of the all-objective protocol, the detection results of each link in the bottom and upper layers, opens the original image data, facilitates the secondary development of partners and algorithm integration, and opens the internal testing and fusion development tool chain to facilitate partner tuning. Trial and two development.
Maxieye can provide location elements and more accurate lane recognition for customized customers of automatic driving.
For low-speed parking cars, Maxieye can provide high-precision target recognition, which focuses on the road edge, deceleration belt and lane line recognition.
For lidar customers, Maxieye can provide full protocol target output, full protocol world coordinate output and full protocol angular coordinate output. With such output, partners can cluster point clouds well, reduce the number of laser lines and reduce the system cost.
For map partners, Maxieye can provide cubic curve equation in local coordinate system of lane vehicle, cubic equation curve in local coordinate system of roadside vehicle, coordinate equation in local coordinate system of parking line vehicle, driving scene category (tunnel entrance, tunnel exit, tunnel inside, judging day and night) . Map manufacturers can superimpose GPS absolute coordinates on the basis of Maxieye's information to generate the absolute coordinates of the target, thus generating the map.
For EPS actuator partners, Maxieye can provide them with cubic curve equation of lane line, category of lane line, farthest credible distance of lane line, bifurcation point of lane and information of adjacent lane.
For chassis actuator partners, Maxieye can provide full protocol vehicle information (relative distance, relative speed), full protocol pedestrian information (relative distance, relative speed), full protocol rider information (relative distance, relative speed).
For intelligent lighting system, Maxieye can provide the output of all-protocol target world coordinates, all-protocol target angle coordinates, street lamp target coordinates, tunnel entrance/exit coordinates.
Zhou Shengyin believes that opening up is one of the means to promote the prosperity of the industry. Maxieye has stepped out of the first mile of opening up, and then how to make the open ecology more prosperous and profitable, which is the problem that Maxieye, a young company, has to face.