Research On Key Technologies In Unmanned Vehicle Driving Environment Modelling Based On 3D Lidar

As the next blue ocean which will change people’s live ways, the research on unmanned vehicle is drawing high attention from the academics and industry, whose development is the combination of cognitive science, artificial intelligence, control science and mechanic engineering. Moreover, the development of unmanned vehicle not only provides an ideal cross validation platform for the new technologies development in these subjects, but also is the inexorable trend for the development of future vehicle.In the research area of unmanned vehicle, how to model the 3D environment completely, accurately, real-time and robustly has been the key and difficult point in research. Velodyne LIDAR has been widely applied in 3D modelling of the environment for its ability of obtaining 3D environment information in the way of non-contact. The technology for the model of 3D environment and the construction of Road Priority Space-Time Situation Map (RPSTSM) based on Velodyne LIDAR is presented in this work. The detailed research content can be listed in following four aspects:1. The relative concepts and the state-of-art concerned on the environment modelling methods of unmanned vehicle is presented systematically. The comparison of advantages between 3D LIDAR and other sensors in 3D modelling is presented, followed by the emphasized introduction of mainstream modelling methods concerned on each elements in driving environment based on 3D LIDAR.2. Study on ground segmentation method of source data generated by 3D LIDAR is presented. Aiming at the fact that the pointcloud shape of source data generated by 3D LIDAR varies in different environments, a multi-feature, loose-threshold, varied-scope ground segmentation method is studied to realize the algorithm’s robustness and accuracy in ground segmentation, based on which obstacle map is built.3. Study on road curb detection based on obstacle map is presented. A road curb extraction method based on the analysis of the road shape is presented, which reduced the influence of obstacle inside the road area and the interruption of the road curb by obtaining road trend and width distribution information. The prediction of the road curb is mainly based on the vehicle’s position and orientation while the updating is mainly based on the height difference map generated by source point cloud data to enhance the real-time performance and continuity.4. Study on modelling of dynamic obstacle and generation of RPSTSM. To improve the extraction integrality of the dynamic obstacle, a dynamic obstacle extraction method based on the neighborhood distribution information of scan line is presented. Based on the extraction result, a shape analyzation method of the dynamic obstacle is performed using the PCA algorithm, which is fused with multi-obstacles’distribution pattern based on quardtree to increase the mapping precision under crowded traffic environment. Whereafter the Kalman filter is performed on the dynamic obstacle to predict the future distribution of obstcles to construct the space-time obstacle map, where the distance transformation is performed on to build RPSTSM. Based on the fusion of various environment elements, RPSTSM reflect the region’s safety risk, which provide a concise and accurate interface for vehicle’s decision.The algorithm presented in this work is tested on "Intelligent Pioneer" and mass of tests are performed. The experiments shows that a complete, accurate, robust, real-time performance is obtained by the presented algorithm. The main contribution is summarized and future work is outlooked at last.