Research On The Detection Of Driving Area Of Intelligent Vehicle Based On Multi Lidar Data Fusion

Lidar has a wide range of applications in the field of perception due to its rich sensing information、long detection range、and low vulnerability to autonomous driving technology of environmental interference.With the increase in the complexity of smart vehicle use scenarios,With the increase of the complexity of the intelligent vehicle use scene,the research on the application of lidar is also continuous in-depth.In order to more fully perceive the surrounding environment of the intelligent vehicle,this thesis has studied the detection of the drivable area based on the method of multi-lidar data fusion,and the specific work content will be expanded from the following aspects:(1)The lidar C32-C and C16-A are selected as the sensing sensors of this study,and their performances are compared and analyzed theoretically.Combined with the specific requirements of intelligent vehicle for environmental perception,the two lidars are reasonably arranged,and C32-C is the primary-lidar and C16-A is the sub-lidar.In order to realize the data fusion of the two lidars,Firstly the external parameters of the primary-lidar are calibrated by the improved chessboard calibration board.Then this thesis proposes a multi-lidar external parameter calibration method.The calibration geometry is placed in the overlap area of the two lidars’ field of view,and the translation and rotation matrix of the sub-lidar relative to the primary-lidar is solved by using the plane vector method,and the external parameter calibration of the two lidars is realized.For the time alignment of the two lidars,the same external clock is introduced,the structure of the two lidar data packets is analyzed,the absolute time of each data block in each packet is calculated,and the data block with the closest absolute time of the two lidars is found,which is the starting point of data storage to achieve time alignment.(2)In this thesis,a multi-step filtering method is proposed to filter outliers,noise points and occluded points in the field of vision of the sub lidar,Finally the improved grid method is used to compress the point cloud and complete the data preprocessing of the point cloud.Then,the double threshold ray slope method is used to segment the ground points,and the radius filtering method is used to filter the incomplete segmentation ground points,so as to achieve stable segmenting effect.Then this thesis proposes an improved OPTICS algorithm,which can reduce the time complexity and ensure the clustering effect.Finally,the similarity evaluation system is constructed according to the characteristics of obstacles,and the Hungarian algorithm is used to achieve the obstacle matching between the front and back frames.In order to find the best matching among the maximum matching results,theevaluation standard of the best matching relationship is constructed,the best matching relationship is selected,and the motion state of the obstacles is estimated by the extended Kalman filter.In order to find the best match in the maximum matching results,the evaluation criteria of the best matching relationship is constructed,the best matching relationship is selected,and the moving state of the obstacle is estimated by the extended Kalman filter.(3)Based on the analysis of the distribution characteristics of point clouds on the structured road,a two-step screening method based on the curvature mutation characteristics and line characteristics of point cloud is proposed.Using the least square quadratic curve to fit the edge of the road.The driving area of intelligent vehicle is constructed along the side line.Finally,the software system of lidar point clouds processing based on ROS platform is developed,and the experiment is carried out by combining with the intelligent vehicle experiment platform of our school.