Research On Environment Perception Algorithm Of Autonomous Electric Formula Race Car

The Formula Student Autonomous China is a series of Formula SAE competitions,which aims to cultivate driverless technical talents.The core technologies of the driverless formula car mainly include environment perception,path planning and vehicle control.This paper focuses on the research of environment perception technology in driverless formula car in order to provide necessary prerequisites for path planning of the car.This project relies on the National Natural Science Foundation of China(51675257)and key research and development plan project of Liaoning province(2017106020)to design a buckets spatial location detection and color recognition algorithm applied to the racing field environment,and at the same time build the environmental racing field map and realize the real-time positioning of the car.For the space position detection of buckets,Firstly,the original point cloud data of Li DAR is set to preserve the effective ROI around the car by setting appropriate threshold limits on the coordinate axis,Secondly,local area point sets are selected to fit the ground plane model to filter ground point cloud data.Then,the Euclidean distance clustering algorithm with adaptive search radius threshold is applied to cluster the obstacle point cloud.Finally,the post-optimization algorithm is used to realize the spatial position detection of buckets.For the color recognition,YOLOv3,the current mainstream deep learning framework,is adopted to collect nearly a thousand image data sets of buckets,label the data sets according to the color information,use the server to acquire the weight file applied to the color recognition of buckets through training,and call the weight file obtained through training in real time to complete the color recognition of buckets.For the data fusion,First,according to the data receiving time,the low-frequency Li DAR data is interpolated to the high-frequency camera data,and the nearest neighbor principle is adopted to achieve timestamp matching;then the camera internal parameter matrix is calibrated,the image collected by the camera is de-distorted,and the direct linear transformation algorithm(DLT)is used to calculate the external parameter transformation matrix,and finally realize the fusion of the three-dimensional space information and color information of buckets through the internal and external parameter transformation matrix.For the positioning of the car,First of all,the inertial measurement unit(IMU)data is used to remove the original point cloud motion distortion of the Li DAR,and then the feature point set is divided by the smoothness information,and the different feature point sets are registered respectively to obtain the front-end odometry,and then the normal distribution transformation algorithm(NDT)Construct the loop constraint,and finally use the graph optimization algorithm to calculate the Li DAR pose,and combine the Li DAR external parameter matrix to obtain the car pose in real time to achieve precise positioning of the car.Simulation verification and competition results show that the algorithm studied in this paper can make the cars complete the sensing and positioning function in the field environment excellently,the accuracy and frame rate of the algorithm can meet the requirements of the competition,and the algorithm won the third place in the FSAC Competition in 2020.