Abstract: Aiming at the obstacle detection problem in environmental perception of unmanned vehicle, a target recognition and tracking method based on onboard lidar is designed. For reducing computation efforts and increasing processing speed, point-cloud filtering and segmentation algorithms are introduced to reduce original laser-point-cloud data, effectively enhancing the real-time performance of detection. Based on SVM classifier, multi-feature compound criteria are used to improve Adaboost algorithm, and three-dimensional point-cloud data are directly processed, retaining perceptual information to the maximum extent and enhancing recognition accuracy. A data correlation method based on maximum entropy fuzzy clustering and corresponding particle filter are proposed to effectively enhance the stability and accuracy of target tracking in complex traffic flow. The data set simulation on Baidu Apollo platform, the experimental verification on self-developed unmanned driving platform and real vehicle verification in small target overlapping and occluding conditions show that the method proposed has good robustness and real-time performance

Key words: unmanned vehicle, environmental perception, lidar, recognition, tracking