| In recent years,with the development of artificial intelligence technology,the application of unmanned driving is increasingly extensive.As an important part of driverless,environmental perception technology mainly includes the information of road surface,obstacles and the determination of the area where driverless vehicles can pass,which is a significant guarantee for the safety and intelligence of driverless vehicles.With the development of sensor equipment,lidar acquisition technology has been widely used in automatic driving,which makes the unmanned vehicle capture more road environmental information and more conducive to the perception of surrounding road conditions.Thus,driverless vehicles can make reasonable path planning and behavior operation.However,due to the high cost of lidar equipment and the complexity of scene data,the research of lidar environmental sensing technology is still a challenging frontier topic.Therefore,our paper focuses on several key technologies of road environment perception of unmanned vehicle based on the point cloud data collected by vehicle lidar,and mainly completes the following works:(1)A ground segmentation method based on piecewise calibration RANSAC is presented.Firstly the scene data is equidistant segmented and the horizontal calibration method is implemented in order to solve the problem of ground unevenness.Then,RANSAC is used to extract the plane information from each piecewise dataset.Finally,the segmentation results of each piecewise dataset are combined to complete the ground segmentation.(2)A voxelized DBSCAN clustering algorithm and obstacle recognition method based on structural features are proposed.In this method,non-ground point cloud data is rasterized to eliminate grid elements that are not conducive to clustering,then DBSCAN clustering is carried out on the centroid points of all dataset in the grid to realize the segmentation of obstacles.According to the segmented point cloud,the structural feature information of obstacles is selected and extracted,and the JointBoost classifier is used to realize feature judgment and obtain the preliminary classification probability.Then the results are optimized through combining the features and the classifier to complete the classification and recognition of obstacles.(3)A road boundary detection method is designed which is based on structural analysis of roads.Firstly,the method extracts the candidate points of road boundary by adopting the abrupt elevation features in point cloud data.Then,the extracted boundary candidate points are screened through the prior constraints obtained from ground segmentation,and the wrong points that fall on the non-road boundary in the road driving area are eliminated.Finally,the correct candidate points are fitted by the least square method to obtain the final road boundary line.(4)An analysis method of road passable area is constructed.Firstly,the data of the road area is raster processed,and the eight neighborhood marking algorithm is adopted to mark the data amount of each grid to determine whether the grid is a passable area,and then the passable area is updated by judging the suspended obstacles on the road boundary to complete the detection of the passable area of the road.Experimental results show that the environmental perception technology implemented in this paper can meet the requirements of real-time and robustness of unmanned driving,and it also enriches the method system of environmental perception technology. |
