Car cloud network 2021-08-31 22:24:58 阅读数:270
In the near future , Qingzhou Zhihang, a domestic autopilot company, released the latest R & D achievements related to lidar sensing technology ——SimTrack, The integrated detection and tracking of multiple target objects can be realized in the point cloud data of lidar （joint detection and tracking）. This research breakthrough relies on “ Track after detection ”（tracking-by-detection） Traditional process methods , It reduces the manually designed matching rules and the related parameter adjustment links , And get very good evaluation results on multiple data sets . This research result （Exploring Simple 3D Multi-Object Tracking for Autonomous Driving） Selected ICCV 2021（IEEE/CVF International Conference on Computer Vision, One of the three top conferences on computer vision ）.
2021 year , Many new models in China have announced plans to carry lidar , Including Wei Lai ET7、 Xiao peng P5、 saic R automobile ES33、 Zhiji L7、 Alpha the fox S、 Which zha S wait , The first year of lidar boarding is about to start . The technical considerations behind , Lies in what is currently available L2-L3 Most of the level assisted driving functions are “ camera + Millimeter wave radar ” Solutions for , Mainly rely on visual perception , The detection accuracy of other traffic participants is limited , There are safety risks . If car companies want to launch a more reliable solution , You can't do without lidar .
Lidar has higher resolution 、 A more stable , And the collected point cloud data is 3D, Than the monocular camera 2D The data has the dimension of depth , It can greatly improve the accuracy of object detection and tracking . therefore , To maximize the safety of automatic driving function , You need to start from the lidar 3D Accurately detect and track all kinds of traffic participants in point cloud data .
In the past, tracking mainly relied on “tracking-by-detection” A set of process methods , However, this kind of method requires manual design of matching rules and involves a lot of manual parameter adjustment , Low deployment efficiency and scalability . To optimize this process , Qingzhou Zhihang put forward SimTrack This algorithm , With the help of an end-to-end trainable model , It can directly detect and track multiple target objects in point cloud data , It greatly reduces the time investment of Engineers . The method can be simply summarized as predicting the position of the object for the first time in each LIDAR point cloud , In order to obtain the tracking identity of each object （tracking identity）, Then with the help of the prediction of object motion , Continuously update the position of each object .
SimTrack It was built by Qingzhou Zhihang “ Autonomous super factory ” An important embodiment of this technical idea . Since its inception , Qingzhou Zhihang attaches great importance to the construction of bottom technical ability , Hope to build a highly automated “ Autonomous super factory ”, To promote the large-scale landing of driverless vehicles .
before , In order to achieve the goal of high automation , Qingzhou Zhihang has pioneered the application of self supervised learning in the sensing direction of lidar , Reduce the dependence on manual annotation ; meanwhile , Traditional robot technology and machine learning technology are combined in planning control technology , It not only enhances the interpretability of artificial intelligence technology , It also reduces the operation of repeated manual parameter adjustment by engineers , Convert engineer time into machine time .
stay “ Autonomous super factory ” Under the support of , at present , Qingzhou Zhihang has been based on Robotaxi The technology stack has been deployed with several Robobus, Surpass in the world 8 Cities carry out operation and testing , More than... Models have been deployed 8 paragraph , Can cope with rainy days 、 Downtown and other complex scenes .
Next , Qingzhou Zhihang will continue to launch Dragon Boat Series driverless vehicles , Applied to urban public transportation 、 Online travel and other scenes , Boost urban transportation towards intelligence 、 Networking 、 Sharing . Besides , Qingzhou Zhihang is still working for the main engine factory 、 Scientific research institutes and other institutions provide a tool chain for automatic driving technology research and development , Help partners to build their own automated data closed loop .
版权声明：本文为[Car cloud network]所创，转载请带上原文链接，感谢。 https://car.inotgo.com/2021/08/20210826121459627m.html