Map building of non-structural environmental characteristics based on dynamic model-hierarchical clustering algorithm

Unmanned Ground Platform (UGP) is a fundamental part of future transportation and military system. However, it still faces lots of difficulties in obtaining a real-time and precise understanding on non-structured natural environment. In this paper, to assist UGP for a better perception of non-structured natural environment, we propose a dynamical clustering algorithm based on model-hierarchy to construct the real map from non-structured environment features. The proposed algorithm first classifies data points based on a simplified model of environment features. Then, various features are extracted for each class. Finally, a dynamic hierarchical clustering algorithm is utilized where an evaluation metric is introduced to control the cluster number automatically. The proposed algorithm can not only eliminate the influences brought by 3D laser point cloud, but also retain the advantages of classic CURE algorithm that can recognize non-balled area. Moreover, this algorithm is self-adaptive to irregular terrace features and independent of settings for initial number of clusters. In the experimental section, the algorithm is tested on the point cloud data obtained from 3D laser radar. The result demonstrates the effectiveness and feasibility of the algorithm.