Using Depth Information for Real-Time Face Detection

초록

This paper proposes a method for real-time face detection. The proposed method is to use depth information to detect faces in a manner that performs robustly even in response to changes of lighting and face sizes. And we use the depth difference features in depth image and apply this to the boosting algorithm for training and recognition. The depth difference features that this study proposes to use enable dynamic changes to be made to the window according to the face size. The conventional method of face detection was to define the size of the face to be detected in advance, and complete scanning for all the possible sizes. By contrast, this new method requires only a single scan to detect faces of all sizes, since it uses the central depth value to predict size based on 2ndpolynomial model. These detected depth difference features are performed training and recognition step with boosting algorithm[1]. The boosting classifier performs recognition by connecting strong classifiers that are constituted by weak classifiers. © Springer International Publishing Switzerland 2014.