| In recent years,with the frequent occurrence of traffic safety accidents and the rapid development of modern intelligent technology,intelligent vehicle has become a research hotspot in the automotive field.Intelligent vehicles need to use safe assisted driving technology to achieve autonomous driving,and the obstacle detection technology in front of intelligent vehicle is the basic premise of safe assisted driving.This topic studies the obstacle detection technology in front of intelligent vehicle,which has high theoretical significance and engineering value.Based on the electric vehicle as the carrier,this paper researches the detection technology of obstacles in front of the vehicle based on the fusion of lidar and machine vision.The main research contents are as follows:(1)Analyze and compare the performance of each sensor in the intelligent vehicle obstacle detection system at present,and design a set of intelligent vehicle front obstacle detection scheme based on the fusion of lidar and machine vision,combined with the structured road scene.The scheme determines the model of the sensor and its installation location.(2)The method of detecting obstacles by lidar is studied.After the calibration of the lidar is completed,the road boundary constraint and the double feature of the height feature and the smoothness feature are used to screen out the suspected points of the road boundary,and the RANSAC algorithm is used to fit these points to realize the extraction of the road boundary.For the point cloud within the road boundary line,a ground plane-based fitting algorithm is used to divide it into ground and non-ground point clouds.An improved DBSCAN algorithm is proposed for non-ground point cloud clustering,and a Bounding Box which characterizes the obstacle contour is generated according to the clustering result.The recognition and detection of obstacles by lidar is realized and verified by experiments.The effect of the algorithm.(3)The obstacle detection method based on lidar and machine vision fusion is studied.Through the conversion relationship between the world coordinate system,the image pixel coordinate system and the lidar coordinate system,the projection of the Bounding Box generated by the point cloud cluster after the lidar clustering is realized,which generated a region of interest on the image.By extracting the Haar-like features and HOG features of a large number of training set samples,a classifier cascaded by each strong classifier is constructed by combining AdaBoost algorithm.The region of interest of the image is input into the cascade classifier for classification and identification of the obstacle.The comparison experiment is carried out with the classification results of purely using lidar detection and the original image placed in the cascade classifier.The experimental results show that the obstacles in the region of interest are classified faster and with higher accuracy.The method realizes redundant detection of lidar and machine vision,and improves the reliability and accuracy of obstacle detection. |
