2. Key Laboratory of Opto-Electronics Information Technology of Ministry of Education, Tianjin University, Tianjin 300072, China
Overview: Obstacle detection is one of important research fields of intelligent vehicle environment perception technology. It is important for vehicles driving to detect the obstacles quickly and accurately. There are two main types of obstacle detection methods: based on visual sensors and based on LiDAR sensors. Since the latter method has the characteristics of no-susceptible to environmental impact, strong anti-interference, high ranging accuracy and etc, it is widely studied and applied in obstacle detection. Cluster analysis is one of the most commonly methods in LiDAR detection. Among them, DBSCAN algorithm is widely used because it can obtain clusters of arbitrary shape without knowing the number of classes in advance and can also identify noise points effectively. In order to detect obstacles quickly and accurately, this paper proposed an optimized DBSCAN algorithm which improves the adaptability under different distance by optimize the selection method of neighborhood radius. The procedure of obstacle detection in this paper includes four steps: road boundary detection, ROI region data extraction, ground data removal and optimized DBSCAN algorithm clustering. Firstly, use the characteristic that the structured road boundary point has obvious elevation mutation than the ground point, detect the local Z-value abrupt changing point and use the least square method to fit out the road boundary. Then, according to road boundary, extract the data of the inside area (the ROI area) of the road boundary. Next, fit the ground plane in ROI area and remove them from ROI. Finally, use optimized DBSCAN algorithm to handle the data in ROI after boundary detecting and removing. The Ford Campus dataset which is acquired by the University of Michigan and Ford Motor Company is used to test the performance of the optimized DBSCAN algorithm. The experiments were performed on a computer with 4 GB memory and 3 GHz clock frequency, and programmed on MATLAB. The experiment results show that the effect of the optimized DBSCAN algorithm is significantly improved for both short-range and long-range targets. Compared with the traditional DBSCAN algorithm, the positive detection rate of obstacle detection improves and the false detection rate reduced significantly. Since some false detections caused by road boundary detection error, we can improve the accuracy of boundary detection by multi-sensors fusion in the future. Considering the driving environment of the unmanned vehicle, multi-sensors fusion can be applied to the algorithm, and the robustness and stability of the algorithm will be further improved.