Image based estimation of pedestrian orientation for improving path prediction

Abstract

Pedestrian protection is an essential component of driver assistance systems. A pedestrian protection system should be able to predict the possibility of collision after detecting the pedestrian, and it is important to consider all the cues available in order to make that prediction. The direction in which the pedestrian is facing is one such cue that could be used in predicting where the pedestrian may move in future. This paper describes a novel approach to determine the pedestrianpsilas orientation using Support Vector Machine (SVM) based scheme. Instead of providing a hard decision, this scheme estimates the discrete probability distribution of the orientation. A Hidden Markov Model (HMM) is used to model the transitions between orientations over time and the orientation probabilities are integrated over time to get a more reliable estimate of orientation. Experiments showing the performance of estimating orientations are described to show the promise of the approach.