Context-based robust face detection algorithm for surveillance cameras (パターン認識・メディア理解)

概要

論文の詳細を見る
This paper describes a context-based robust face detection framework for surveillance cameras. Different from familiar faces in our daily lives, faces captured by surveillance cameras are smaller and darker with motion-blurs and distortions. Furthermore from cameras up there, e.g. on ceilings, faces are downward and partially unseen from cameras. To detect such varied and degraded faces, we utilize contextual information about faces of walking people in surveillance cameras. We built a probabilistic detection framework combining a face detector with contextual information. Firstly we use a boosted face detector to calculate a primary probability distribution of possible face regions. After this fast filtering to select small amount of possible face regions, we use a HoG feature-based outline detector to calculate a conditional probability from neighboring regions. Combining those two detector-based probabilities with probability of face sizes estimated from a camera configuration, we achieved a high face detection rate of 93.7% with about 1,000 times lower false positive rate than one in the case of only face detector as well as keeping computational efficiency of the boosted face detector.
社団法人電子情報通信学会の論文
2008-12-11