Research On Environment Perception Technology Of Self-Driving Vehicle In Urban Environment

In recent years,with the development of computer technology and the reduction of sensor cost,significant progress has been made in the research of autonomous driving system,and the technology of self-driving car is becoming increasingly mature.Selfdriving car have the potential to improve safety,productivity,and the usage rate of road,as well as good for the environment.Self-driving cars will become the main transportation in urban transportation systems in the future.The autonomous driving system is generally composed of several parts,such as environment perception,path planning and vehicle control.environment perception refers to the ability of the self-driving car to understand the environment,such as obtaining the location information of the obstacles in the environment and detecting the traffic signs in the environment.Environmental perception is the basis of route planning and vehicle control and is also the most important part of the whole autopilot system.Compared with other road conditions,the road condition in urban environment is more complex and diverse,requiring more environmental information perception,such as kerb and traffic sign detection.Therefore,this paper mainly studies the environmental perception technology of self-driving cars in urban environment,designs and realizes multiple perception tasks of self-driving cars in urban environment.The specific contents are as follows:Firstly,a Lidar based kerb detection algorithm is designed based on the structural characteristics of urban roads.In the first place,this algorithm simplifies the preprocessing process of lidar data and use a sliding window algorithm to extract rough kerb points.Then,use the kerb position predicting method to process these rough points to get the exact position of the kerb.This algorithm not only guarantees the operation efficiency,but also effectively improves the detection accuracy.Secondly,this paper uses deep learning technology to design a camera-based traffic sign detection method,which can effectively detect common traffic signs in daily driving.The method is combined with optical character recognition technology to recognize the speed limit information in the speed limit sign.Finally,this paper presents a vehicle distance detection method based on multisensor.Firstly,the depth information is extracted from the LiDAR data by using the deep neural network designed in this paper.Secondly,the extracted depth information is matched with the two-dimensional vehicle information extracted from the camera data by the two-dimensional target detector,and finally the distance information of the vehicle is obtained.