Study On Terrain Estimation And Attribute Cognition Based On Laser Scanning And Vision

The Multi-sensor data fusion and environmental terrain estimation constitute important foundations for autonomous navigation of Unmanned Ground Vehicles(UGV).The thesis explores the method of environmental terrain estimation based on Radial Basis Function. By combining non-stationary kernel template with terrain grid, the research accommodates the non-uniform density distribution of laser point cloud with the increasing detection range and reduces the negative impact of abrupt shafts on ground, and in a result provides explicit topographic maps for autonomous navigation. Treating the point cloud in layers deals with impending obstacles so that the estimation of culverts and trees becomes more accurate. In addition, treating the original point cloud with VoxelGrid filter rationalizes the distribution of point cloud and improves the efficiency of calculation.For the application environment of UGV, based on the sensing system composed by in-car 3D lidar, Omni-vision and GPS-INS, the thesis uses the location and attitude of UGV that GPS-INS provides to align point cloud, which offsets the impact of UGV attitude change and compensate most of the blind zone in single-frame point cloud. The thesis also combines environmental color information that the Omni-vision sensor returns with location and depth information of 3D point cloud for fuzzy classification by color saturation, hue, possibility of ground points and surface roughness, which realizes the estimation of terrain feature.In the last part, the author does simulation experiments in an interactive system, which is set up with Robot Operating System and a model of UGV system. Along with some field experiments, all terrain estimation approaches get verified.