Bayesian sequential face detection with automatic re-initialization

摘要: This paper proposes a probabilistic search algorithm to boost the computational efficiency of face detection in video sequences. The sequentially predicts probability distributions region parameters current frame given sequence past frames, and automatically re initializes prediction process at scene changes sequence. A Bayesian criterion is derived as determinant likely regions among collection candidates generated by sub window-based classifier. scheme also enable marginalization multiple outputs Experimental results on test 500 frames broadcast video, containing 450 faces, demonstrate that proposed approach achieves speed roughly double baseline detector included open-source computer vision library OpenCV, without sacrificing performance.