Research On Vehicle Recognition Based On Fusion Of Vision And Radar Data

With the rapid development of driverless technology,environmental sensing,as one of the key technologies of driverless,is to acquire target objects by combining sensor and vehicle control center.This paper is based on the project of Science and Technology Platform of Liaoning Education Department(JP2017006),which uses lidar sensor and visual camera to collect data,and obtains the required data through the Data pre-processing,the MV3 D data fusion algorithm is used to detect the flat road objects,and the algorithm is validated.In order to improve the efficiency of target detection in the fusion process,the YOLOv3 algorithm is used to train and classify the target.Details of the study are as follows:(1)Data fusion is analyzed generally,including the system composition of multi-sensor,image data filtering and de-noise processing,laser radar point cloud data processing and other pre-processing operations,so as to lay a foundation for subsequent multi-sensor data fusion.(2)YOLOv3 target detection algorithm is used in this paper.Through the training of the algorithm,we can identify the kind of target and detect the object,and verify the algorithm,which can ensure the real-time accuracy of the vehicle detection in front.(3)The calibration of the inner and outer parameters of the vision sensor and the joint calibration of the multi-sensors can realize the correlation transformation of the radar coordinate system and the pixel coordinate system,the calibration of the radar point cloud data,and the time and space fusion of the laser radar and the camera,and the coordinate correction after the fusion is carried out to ensure that the time stamp is consistent.MV3 D target detection is used as the multi-sensor fusion framework in this paper,and the pre-processed data are input into the fusion algorithm for vehicle distance and position identification.In this method,the low-dimensional map,that is,the bird’s-eye view of the point cloud,is represented by a highly accurate 3D candidate box.(4)Build a vehicle verification platform for vehicle verification analysis.In this paper,based on the modified Chinese V3 experimental vehicle,the pilotless experimental platform is constructed,and the position of the lidar sensor and the vision sensor is adjusted and optimized.Firstly,the data packet is collected on the main road section of the campus,and the real-time detection performance of the fusion algorithm is tested and analyzed in the 9th teaching building under the moving object condition by using the data packet information,by comparing the detection distance between the measurement tool and the algorithm,it is ready for the outdoor vehicle verification and analysis.(5)Carry on the real vehicle verification to the algorithm which completes the debugging,this article sets many kinds of experimental working conditions according to the national standard to carry on the examination to the vehicle in front,in the experimental test process,compared with single sensor,the fusion scheme has higher detection speed and accuracy,and can detect the target and distance of the vehicle in front in real-time,and the detection result is effective.