Fast Face Detection with Precise Pose Estimation

Francois Fleuret (INRIA, France)

We present a fast algorithm for face detection and precise pose estimation. Our detection scheme is based on a tree-structured hierarchy of face vs. background classifiers combined with a lazy evaluation strategy which concentrates computation on ambiguous areas of the image. The hierarchy corresponds to successive partitions of the pose parameter space, and thereby provides, at no additional cost, a fine estimate of the 2D pose of detected faces.