Research On Obstacle Detection Technology Of Vehicle-borne Lidar In Unstructured Environment

With the progress of science and technology,intelligent unmanned technology has been developed rapidly,especially in the field of driverless vehicle successfully landing.However,the current driverless technology is mainly oriented to urban,highway and other structured environment,but the development of driverless technology in farmland,grassland,gobi and other unstructured environment is slow,agricultural production,border patrol and other work still need to invest a lot of manpower.Accurate and robust obstacle detection is the cornerstone of unmanned driving.However,the ground in unstructured environment is more undulating,the interference factors are more diverse,and the obstacles are more irregular,which makes the accuracy and robustness of obstacle detection in unstructured environment lower.Aiming at the problem of low accuracy and robustness of obstacle detection in unstructured environment,this paper studies the obstacle detection technology of vehicular lidar in unstructured environment.This paper studies the installation layout,time axis synchronization,space coordinate system I,point cloud motion distortion compensation,environmental noise removal,ground segmentation and positive and negative obstacle detection of two lidars on unmanned vehicles,and carries out real vehicle experiments in real scenes through self built experimental vehicles.The main research contents are as follows:(1)Aiming at the complex structural characteristics of unstructured environment,the overall design and experimental system construction are studied.The main problems of obstacle detection in unstructured environment are analyzed,and the overall design scheme is determined;the experimental vehicle is built,and the installation layout scheme of dual lidar suitable for unstructured environment is proposed,and the joint debugging of hardware modules and software systems of vehicle sensing system is completed.(2)Aiming at the problem that the space-time coordinate system of vehicle borne dual lidar is not unified,the research on the synchronization of time axis and space coordinate system of vehicle borne dual lidar is carried out.The vehicle GPS is used to trigger the two lidars for unified time service,so that the point cloud data of the two lidars have a unified UTC time stamp standard,and the two lidars can send data frames synchronously;the preliminary spatial coordinate system of the two lidars is established through the radar layout scheme designed when the experimental system is built,and on this basis,the ICP point cloud fine registration algorithm is used The precise coordinate system of two lidars is established.(3)Aiming at the problem of point cloud motion distortion and environmental noise in unstructured environment,the research on point cloud motion distortion compensation and environmental noise removal is carried out.The motion distortion compensation method of LIDAR point cloud based on six axis IMU is designed.The built-in IMU of lidar is used to compensate the rotation of point cloud according to the idea of data block partition proposed in this paper.The inter frame motion of point cloud is estimated by voxel filter,generalized ICP,extended Kalman filter and IMU information,and the inter frame motion is used to compensate the translation of point cloud;Combined with the divergence angle of laser radar beam,an adaptive DBSCAN density clustering algorithm is proposed to remove the environmental noise.(4)Aiming at the low accuracy and robustness of obstacle detection in unstructured environment,the research on obstacle detection in unstructured environment is carried out.Firstly,the undulating ground is segmented by the improved ray method;secondly,the positive obstacles are detected by the height difference between the positive obstacles and the ground point cloud;thirdly,the local geometric features of the negative obstacles are analyzed,and the negative obstacles detection method based on the local geometric features is designed.Real vehicle experiments in multiple scenes show that the improved ray segmentation algorithm performs better in segmentation of undulating ground.The minimum stable detection distance of negative obstacles is 8m,the successful detection rate of positive obstacles is not less than 90%,and the total time consumption of ground segmentation and positive and negative obstacles detection algorithm is less than 50 ms.In addition,real vehicle experiments in multiple scenarios also show that the proposed method has strong robustness.