Abstract: Pedestrian recognition in cross-country environment is a fundamental requirement of group accompanied automatic driving vehicle. Aiming at the problem of pedestrian recognition in lidar point cloud data, especially under cross-country environment, a clustering-based solution is proposed in this paper. On the basis of theoretical analysis, combined with the geometric and physical features of human being, a pedestrian recognition algorithm based on KDTree and Euclidean clustering is designed and a corresponding test is carried out on tracked vehicle under corss-country environment. The results show that the lidar-based pedestrian recognition algorithm designed can accurately identify the pedestrians in lidar point cloud data, with a good recognition rate in cross-country environment.

Key words: pedestrian recognition, lidar, clustering