Research On Perception Technology With 2D Laser Range Finder For Unmanned Ground Vehicle In Structured Environments

The structured environment perception technologies with 2D laser range finder are researched for Unmanned Ground Vehicle(UGV) in this thesis, including environment model construction, the detection and understanding of the road region, and the moving vehicle detection and tracking. The main results and innovations of the thesis are summarized as follows:1. According to the human visual perception theory, a hierarchical model for outdoor dynamic environment is built. By utilizing the different characteristics of different layers, this model compactly represents the surrounding environment information of UGV. Experimental results show the validity of the model.2. A curb detection method based on the sequential data from the 2D laser range finder has been proposed. It includes three algorithms: a straight curb detection algorithm based on Hough transform with multiple constraints, a curved curb detection algorithm utilizing the 1D Gaussian regression, and a multi-model Random Sample Consensus(RANSAC) curved curb detection algorithm. Experiments demonstrate that this method can accurately and efficiently detect both the straight and curved curbs, and the accuracy rate reaches 86.8%.3. Based on the Markov random field model, the method for extracting the principal direction of the environment is proposed. This method models the problem of detecting the principal direction as a multi-variable optimization problem. The object function only has a global minimum due to the reasonable object function and the range of variables. The convergence rate of the algorithm is accelerated by sparsifying the variables and setting proper initial values. Experimental results show that this method can correctly extract the principal direction of the environment, and the runtime of the algorithm is less than 100 ms, which meets the real-time requirement.4. A moving vehicle detection and tracking method based on the 2D laser range finder has been proposed. This method firstly adopts an improved vehicle measurement model to preliminarily detect and track the vehicle, then the detection and tracking precision is improved by a newly proposed the precise position of the vehicle geometrical model algorithm based on the non-linear constrained optimization problem. A 3D point cloud reconstruction algorithm for the moving vehicle based on a combination of multiple filters has been also built. Experimental results demonstrate that the static or moving UGV can accurately detect and track single or multiple moving vehicles on roads, and realize the 3D point cloud reconstruction.The above mentioned works have been successfully applied to the key project of the major research plan of the Natural Science Foundation of China(NSFC): ‘Research on key scientific problems in intelligent driving of highway vehicles’(90820302).