Pedestrian with direction detection using the combination of decision tree learning and SVM

Abstract

In the real world scenario of automatic navigation of a pedestrian on busy roads is still a challenging job to identify the pose and the direction of the pedestrian. To detect pedestrian with direction will involve large number of categories or class types. This paper proposes a combination of two techniques. The goal is to train the system on the bases of gradients, use the decision tree from that we can generate the candidate list (confusing pairs) with similar features. Taking the confusion matrix into consideration SVM is trained; this method will reduce the computational cost and generate appropriate results. The proposed work can be to classification and track the pedestrian direction on the still images.