Pedestrian identification with distance transform and hierarchical search tree

Abstract

This work develops a novel and robust hierarchical search tree matching algorithm, in which the Distance Transform based pedestrian silhouette template database is constructed for efficient pedestrian identification. The proposed algorithm was implemented and its performance assessed. The proposed method achieved an accuracy of 89% true positive, 92% true negative and low false positive 8% rates when matching 1069 pedestrian objects and 568 non-pedestrian objects. The contributions of this work are twofold. First, a novel pedestrian silhouette database is presented based on the Chamfer Distance Transform. Second, the proposed hierarchical search tree matching strategy utilizing Fuzzy C-means clustering method can be adopted for mapping and locating pedestrian objects with robustness and efficiency. © 2009 Springer Berlin Heidelberg.