Study On Vehicle Detection Based On Fusion Of Image And Lidar

Vehicle detection is one of the key technologies to solve traffic safety and traffic congestion,and vehicle detection is the most basic part of intelligent transportation system.Multi-sensor detection fusion for intelligent vehicles can improve system performance,and effectively provide accurate,reliable and robust information of surrounding vehicles for intelligent vehicles.At present,the vehicle detection algorithm is relatively mature,but the accuracy of multi-sensor detection association needs to be improved under the condition of dense targets.In this thesis,the current mainstream mature detection algorithms are used for realizing vehicle detection in image and lidar.A multi-sensor vehicle detection association algorithm based on coherent point drift matching is proposed,and the correlation results are fused by covariance intersection algorithm.Finally,the effectiveness of the proposed methods are verified by experiments.The main work of this thesis includes the following:1.Firstly,the background and significance of vehicle detection for intelligent vehicles are summarized in this thesis,the research status of vehicle detection and multi-sensor fusion at home and abroad was summarized,the research topic of this thesis was proposed,and elaborated the relevant technologies and algorithms used in this topic.2.An image-oriented vehicle detection algorithm based on Haar and Adaboost is proposed.Firstly,Haar features are extracted from a large number of positive and negative samples,and eigenvalue vectors are established.Then,the weak classifier based on Adaboost algorithm is weighted to form a strong classifier to construct cascade classifier for vehicle detection.A vehicle detection algorithm based on L-type lidar is proposed.Firstly,the point cloud data are pre-processed and clustered to extract the internal features of the cluster,and then matched with L-type features to realizevehicle detection.3.A multi-sensor vehicle detection association algorithm based on coherent point drift matching is proposed.Firstly,the problem of multi-sensor vehicle detection association is regarded as the problem of probability density estimation,in which the image detection results are expressed as the centroids of the Gaussian mixture model,and the lidar detection results are expressed as the data points of the Gaussian mixture model,so the Gaussian mixture model is established.Secondly,expectation maximization algorithm is used t for solving the problem,and the relationship between image and lidar detection results is obtained.Then the correlation results are fused by covariance intersection algorithm to improve the detection accuracy.Finally,a comparative experiment is designed to verify the proposed multi-sensor detection association and fusion algorithm.The experimental results show that the research work in this thesis has certain reference value for future vehicle detection research based on image and lidar fusion.