Obstacle Detection Method Based On Lidar In Port Environment

Obstacle detection is one of the key functions of environment perception for autonomous vehicles,of which the applied strategy is highly relevant to the actual environment.The port traffic scenario is totally different from common urban or highway traffic scenario because of the following aspects:1)road surface in port is very uneven;2)plenty of static obstacles in different kinds of classes could appear anywhere;3)most of the moving objects are semi-trucks which have changing appearance.Though port is one of the excellent scenarios for autonomous driving to achieve real application,few researches till now focused on obstacle detection methods for this scenario.Therefore,we proposed here a Li DAR based obstacle detection method that specifically designed for this scenario.The main research contents are as follows:(1)Preprocessing of 3D Li DAR point cloud,including Li DAR external parameters calibration and interference points filtering.A depth map based dynamic threshold interference point filtering algorithm is proposed,aiming at solving the problem of low efficiency of three-dimensional space search based filtering algorithm.In the rainy weather verification,the proposed method shows better real-time performance than the three-dimensional filtering algorithm,and can preserve more environmental feature point cloud.(2)Ground point cloud segmentation.In order to solve the problem of poor accuracy of ground segmentation algorithm caused by bumps of ego-vehicle,a real-time ground segmentation algorithm that combines model fitting and line scan method is proposed.A horizontal plane fitting step is carried out to calibrate the point cloud and eliminate the impact of ego-vehicle bumps.After that,a scan line is constructed on the calibrated point cloud to perform a slope threshold based ground segmentation method.Experimental results show that the proposed method could realize a real-time and robust result,which has a good performance during driving on both flat and uneven road surfaces.(3)Point cloud segmentation of dynamic and static obstacles.Plenty of static obstacles in different kinds of classes,which is difficult to handle,could appear in port traffic scenario.In order to improve the performance of obstacle clustering step,a separating process of moving and unmoving obstacles is carried out after ground segmentation.To address the unstable segmentation problem of traditional plain grid maps while segmenting dynamic and static obstacles,a spatiotemporal consistent simple grid map based method which uses neighborhood grids and existing frame queue to enrich the global grid map is proposed.In addition,in order to make the detection of large trucks in low speed better,a Bayesian-based grid map method which uses a binary Bayesian filter to fuse local and global grids is proposed.In the test of both single point and practical environment,the proposed methods could not only meet the requirements of real-time,but also achieve a better dynamic and static segmentation performance than traditional methods.Meanwhile,Bayesian-based grid map method could improve the segmentation of moving trucks.(4)Dynamic obstacle point cloud clustering.Traditional grid-based clustering algorithm has poor accuracy of semi-trucks while could realize high real-time.In order to improve the performance of the semi-trucks clustering,referring to DBSCAN,a polar coordinate grid density clustering based on dynamic threshold is proposed.The proposed method uses the number of points contained in each grid as the grid density while taking the dynamic point cloud after dynamic and static segmentation as input.After that,by calculating neighborhood range dynamically,the grid is marked using neighbor’s density as the growth condition.Finally,the dynamic obstacle list is established according to the marked grid.Experimental results show that the proposed method could outperform traditional grid method under the over-segmentation and under-segmentation problems,while both algorithms have satisfactory real-time performance.