Recursive Context Reasoning for Human Detection and Parts Identiication

1 Recursive Context Reasoning for Human Detection and Parts Identi cation Liang Zhao and Chuck Thorpe The Robotics Institute Carnegie Mellon University Pittsburgh, PA 15213 Email: [email protected] Abstract Human detection and body parts identi cation are important and challenging problems in computer vision. High performance human detection depends on reliable contour extraction, but contour extraction is an under constrained problem without the knowledge about the objects to be detected. This paper proposes a recursive context reasoning (RCR) approach to solving the above dilemma. A TRS1-invariant probabilistic model is designed to encode the shapes of the body parts and the context information | the size and spatial relationships between body parts. A Bayesian framework is developed to perform human detection and part identi cation under partial occlusion. A contour reconstruction procedure is introduced to integrate the human model and the identi ed body parts to predict the shapes and locations of the parts missed by the contour detector; the re ned contours are used to reevaluate the likelihood ratio. Therefore, contour extraction, part identi cation, and human detection are improved iteratively. The experimental results of the RCR approach to human detection and body parts identi cation in cluttered scenes are very encouraging.