Design And Implementation Of Environment Perception System For L4 Level Autonomous Driving Vehicle

In recent years,the research and development of autonomous driving in academia and industry has become more and more urgent,environmental awareness technology is one of the core and key technologies of autonomous driving.It is also the cornerstone of the whole autonomous driving system.Therefore,this article focuses on the analysis of environmental awareness module.Autonomous driving vehicle's environment perception relies on sensors such as cameras,Li DAR,millimeter wave radar,ultrasonic radar,camera and so on.Therefore,different scenarios,conditions and weather requires different sensors and different sensor data fusion strategies to achieve the best external environmental perception effect.Among them,L3 level autonomous driving mainly relies on camera,L4 level automatic driving mainly relies on Li DAR besides camera.In this study and this paper,we focus on the development,testing and verification of the key algorithms of camera and Li DAR.In this paper,an improved fully convolution network(i FCN)is proposed for the semantic segmentation algorithm of forward-looking camera to balance the semantic segmentation accuracy and the time consuming of handling each picture.i FCN changes the structure of traditional FCN,i FCN introduces weighted dense jump connection in decode layer.Using adaptive learning rate,set different learning rate for lower and deeper layers.And finally use entropy based feature selection to continuously improve the effect of valuable layer.In this paper,Li DAR point cloud clustering is studied.In order to solve the difficulty of segmentation point cloud on complex road conditions,a method of road point cloud segmentation combining the geometric characteristics of point cloud cluster is proposed.To solve the problem of unstable point cloud clustering,according to the output information like on-board meteorological sensor and humidity sensor,we can dynamically judge the current external environment state,and set different clustering thresholds for different external environment.Experiments show that the research in this paper improves the environment perception ability of intelligent vehicles.