Abstract: In Driver assistance systems the pedestrian protection is an essential component which should be able to predict the probability of collision after detecting the pedestrian and it is also important to consider all the cues available in order to make the prediction. One such cue is the direction in which the pedestrian is facing which could be used in predicting where the pedestrian may move in future. The proposed method describes a novel approach for solving the problem of Pedestrian orientation classification by SVM based Binary Decision Tree (SVMBDT) architecture for solving multiclass problem. The hierarchy of binary decision tree subtasks the use of SVM for decision making. The proposed algorithm takes advantage of both the efficient computation of decision tree and high classification accuracy of SVM. Experiment shows the performance of estimating orientation and describes to show the promise of the approach.
Keywords: Orientation, Multiclass problem, SVMBDT, Binary decision tree.