Construction Of A Test System Of Autonomous Vehicles Environment Perception Algorithm Based On Virtual Modeling

In order to complete the test of the autonomous driving algorithm before boarding the car and solve the problems that may arise in the actual vehicle test,this paper builds an intelligent driving vehicle environment recognition and evaluation system and completes the test of the perception algorithm to improve R&D efficiency and reduce R&D costs.By discovering and reproducing problems in the software simulation environment,no real environment and hardware are needed,which can greatly save cost and time,and speed up model iteration.The main research contents of the thesis are as follows:1)Build a system framework and clarify system functions: This system is a comprehensive system that integrates data collection and labeling,virtual environment modeling,and intelligent driving environment recognition algorithm testing.Complete the selection of system hardware and software platform.The hardware platforms are: high-speed wire-controlled chassis,inertial navigation module,lidar,and color camera;the software platforms are: data labeling platform MATLAB,virtual environment modeling and environment perception algorithm simulation test platform CARLA.2)Based on the Apollo hardware platform,road information(high-precision positioning,3D point cloud and panoramic photos)is obtained in a real environment.3)Use the MATLAB autopilot toolbox to label images and point cloud data(traffic lights,roadsides,intersections,lane lines and other traffic signs,etc.).On the one hand,an intelligent driving environment database can be established for training and simulation of intelligent driving environment perception algorithms;on the other hand,the labeled data can be converted into Open Drive standard road information and sent to the virtual environment modeling platform CARLA to complete virtual scene modeling.The system can use the modeled virtual environment for data enhancement and add it to the intelligent driving environment database.4)After the virtual scene is established,it will provide test scenarios for the environment perception algorithm evaluation platform CARLA,and complete the algorithm test of environment perception in the simulation environment(this article takes the target detection algorithm YOLO,Faster RCNN and the lane line detection algorithm U-Net as examples.test).After comparing the Io U with the real value Ground Truth,it is concluded that the system can be used as an environment perception algorithm for autonomous driving vehicles in virtual scenes.Finally,the evaluated algorithm is iterated and implemented on the hardware platform to form a set of methods for the development and testing of intelligent environment perception algorithms.