Research On Roadside Recognition Algorithm Based On Road Overall Information With Lidar

As an important component of urban structural roads,roadside is insurmountable and an important dividing line between roadable area and unroadable area.The location of the curb can be regarded as one of the most important constraint conditions under the circumstance of changing lanes to overtake and pulling over.Rapid and accurate recognition of roadside information is of great significance to the environmental perception of driverless vehicles.Compared with other sensing sensors,lidar is highly valued by experts and scholars in the field of intelligent driving due to its advantages such as stable performance,strong anti-interference ability and no influence of light.Based on an autonomous brand driverless car platform,this paper studies the lidar roadside recognition algorithm based on the overall road information.The driverless car's lidar is installed in front of the vehicle's intake grid and is easier to identify the curb than the lidar mounted on the roof.The point cloud data collected are converted into spatial cartesian coordinates.Point cloud data elimination method is constructed and preprocessed with filtering method.Then the road point cloud features are extracted from the preprocessed data.The turning point detection algorithm is used to identify the turning point along the road.Finally,the least square cubic curve is used to fit the boundary line of road edge,and the recognition of road edge is realized.The specific research contents of this paper are as follows:(1)Study the lidar point cloud modeling method.The advantages and disadvantages of the lidar in different installation positions are systematically analyzed,and the data are collected by installing the lidar in front of the air intake grille.The distribution of effective road surface point cloud data was studied,and the point cloud data removal method in the three directions of x,y and z was proposed.The isolated points and random noise were filtered successively to obtain the pre-processed point cloud.By comparing the advantages and disadvantages of rightangle grid and circular arc grid modeling,circular arc grid modeling was selected to model the lidar point cloud data.(2)Study the road surface segmentation algorithm of point cloud.This paper analyzes the morphological changes of vehicle-mounted lidar pavement point cloud data and the differences between obstacle point cloud and pavement point cloud,builds two methods to identify the characteristics of pavement point cloud,and selects its threshold value to identify the road point cloud in the experiment.An obstacle point cloud diagram is constructed by identifying the road surface point cloud and the pretreated point cloud.The obstacle point cloud is expanded,and the relationship diagram between the distance and horizontal Angle of the obstacle and the lidar is constructed to obtain road information such as road extension direction and width.(3)Study the roadside extraction algorithm of point cloud.In the obstacle point cloud diagram,the morphological characteristics of the turning point along the road are analyzed,and the point cloud diagram of the intersection area between the road point cloud and the obstacle point cloud is constructed.Combining the road information,the identified inflection points were clustered,and finally the inflection points along different road edges were fitted with the least square cubic curve after clustering.With the vehicle provided by the self-owned brand as the hardware platform and the ROS robot operating system as the software platform,a large number of tests and real vehicle experiments were conducted to verify the feasibility,effectiveness and robustness of the lidar roadside recognition algorithm proposed in this paper based on road information.