A new approach for face detection with omnidirectional sensors

Abstract

This paper tackles the problem of frontal face detection on omnidirectional vision sensors. The adaptation of traditional face detection frameworks for our system enables to speed up the detection process while achieving good performance. We propose a method to synthesize omnidirectional image database. Hence, we were able to train our detector on the synthesized omnidirectional image database. Our experiments suggest that despite non linear distortions, our face detection algorithm has high success rate. We studied performance of Adaboost algorithm variants for our system. The proposed approach is able to match the speed and robustness of existing face detection algorithm for conventional cameras.