Research On Environment Modeling Method And Application Of Multi-sensor Fusion

Driverless technology is considered to be one of the development trends in the future,and it is the hotspot of current research.People hope that driverless technology can alleviate all kinds of traffic problems.Driverless vehicles sense the surrounding environment through various sensors.Multi line lidar and millimeter wave radar are widely used in driverless vehicles.Based on the lidar original point cloud data and millimeter wave radar data,this paper studies the environment modeling technology,and the main research contents are as follows:(1)A point cloud segmentation algorithm based on multi-features of point cloud space neighborhood is proposed.First,the maximum and minimum height difference features in the grid are used to coarsely segment the point cloud,and then the point cloud in the ground grid that is roughly segmented is subjected to secondary processing.Order the point clouds in the ground grid and smooth the point clouds on each laser line with Gaussian filtering,and then combine the curvature characteristics of the point cloud on the two-dimensional plane with the distance characteristics of the adjacent points on the single-beam laser scanning line and the maximum and minimum height difference features of adjacent points on the single-beam laser scanning line segment the ordered point cloud,which can retain more low-low roadside feature points and have fewer false detections.Experimental results show that the algorithm shows high accuracy and robustness in both structured and unstructured scenarios,and it also meets the real-time requirements of unmanned driving.(2)Based on the point cloud segmentation results,the grid method is used to construct the environment model,and the road boundary and road obstacles are detected in the grid map.Use the bidirectional scanning method to obtain the initial roadside candidate points in the grip map,remove the interference during occlusion / gap according to the inherent properties of the road,and supplement with the normal roadside points to obtain the final roadside candidate points,and finally use the RANSAC algorithm fit the road boundary.For the obstacle grids within the boundary of the road,the area growth algorithm is used for cluster detection,and the PCA algorithm is used to obtain the smallest rectangular frame of the target object.Experimental results show that the road detection algorithm in this paper is robust and can accurately detect road obstacles and describe the outline characteristics of the obstacles.(3)This paper proposes a multi-sensor fusion method for moving target detection.First,the stable dynamic targets detected by the millimeter wave radar are selected and the extended Kalman filter algorithm is used to achieve the target's motion state estimation,and then it is fused with the targets detected in the raster map to achieve accurate detection of the moving target and Obtain the accurate outline and motion characteristics of the target.(4)An off-line version of environment modeling system based on lidar data and MMW radar data is designed and implemented.Using the data collected by the unmanned driving platform,the off-line simulation environment is used to debug the code and verify the effectiveness of the experimental results.