Integrated Pedestrian and Direction Classification Using a Random Decision Forest

Junli Tao, Reinhard Klette; Proceedings of the IEEE International Conference on Computer Vision (ICCV) Workshops, 2013, pp. 230-237

For analysing the behaviour of pedestrians in a scene, it is common practice that pedestrian localization, classification, and tracking are conducted consecutively. The direction of a pedestrian, being part of the pose, implies the future path. This paper proposes novel Random Decision Forests (RDFs) to simultaneously classify pedestrians and their directions, without adding an extra module for direction classification to the pedestrian classification module. The proposed algorithm is trained and tested on the TUD multi-view pedestrian and Daimler Mono Pedestrian Benchmark data-sets. The proposed integrated RDF classifiers perform comparable to pedestrian or direction trained separated RDF classifiers. The integrated RDFs yield results comparable to those of state-of-the-art and baseline methods aiming for pedestrian classification or body direction classification, respectively.