Model-based tracking of human walking in monocular image sequences

We present a simple but effective method for recovering a 3D human body, walking parallel to the image plane from monocular video sequences based on a 3D human body model, using robust image matching and a specific search strategy. Monocular 3D body tracking is challenging, because of a large number of free parameters. These parameters are reduced to 12 under the assumption of walking parallel to the image plane. To effectively measure the matching error a pose evaluation function combining both boundary and region information is provided. We simultaneously calculate the physical forces that are used as heuristic information in subsequent search. Experiments with both indoor and outdoor scenes demonstrate that our approach has an encouraging performance.