Research On Technology Of Road Information Extraction Based On Four-layer Laser Radar

With the rapid development of economy and transportation system, intelligent vehicle technology, as an important part of intelligent transportation system(ITS), has been wildly concerned and researched. It is also considered to be the key to solve the traffic problems. Four-layer laser radar has advantages of high accuracy, moderate amount of data, multi-layer detection, less influenced by the environment and so on. It is widely used in the environment perception of intelligent vehicle(IV). In this dissertation, we study the environment detection technology of IV by using four-layer laser radar. This research mainly including the following contents:Firstly, an improved density based spatial clustering of applications with noise clustering algorithm is proposed, it based on the working principle and the data’s distribution characteristics of the laser radar. The data is preprocessed by median filtering and coordinate transformation. According to the data’s distribution characteristics of urban roads and comparing with different cluster methods, the one-dimensional kernel density estimation and 4 nearest neighbors combined weighted Euclidean distance is used to generate the parameters(Eps, Minpts) adaptively and complete the multi-density clustering. The algorithm search the neighborhood of peak area and its searching range is modified by hierarchical processing. The performance of the algorithm is further improved.Secondly, road information extraction method based on the characteristics of structured and semi-structured urban road is proposed. On the basis of improved clustering algorithm and the features of different objects, we obtain the point sets and height of curbs. Then, the point which height is around the curb’s level is marked. At last, the information of obstacles is extracted from the remainder sets, such as distance, position and dimension etc. The methods for extracting the information of curbs, road surface and obstacles are verified by experiments in different scenarios.Thirdly, a modified ICP method is proposed to quickly match the objects of two adjacent frames. Multi-feature matching function is constructed to replace Euclidean distance to improve ICP’s matching strategies. The method could reduce the calculation and the running time greatly.Fourthly, the methods of updating grid maps and tracking moving targets are studied. Grid map is used to describe the front area of IV. The grid maps between two adjacent frames are aligned by the result of matching objects and updated by Bayesian theory. Static and dynamic obstacles are detected by the occupation probability of grid. The dynamic targets on the road are stable tracked by Kalman filter and tracking management.Finally,Beijing University of Technology’s IV, as the experimental platform, is used to do massive experiments, the results show that the proposed methods can stably and accurately extract the information of road and targets.