Model-based human body tracking

Visual tracking of human body movement is a key technology in a number of areas. We present a 2-D model-based method of human body tracking from a monocular video sequence. Morris and Rehg (1998) put forward a 2-D scaled prismatic model for figure registration which has far fewer singularity problems than 3-D models. Here we extend it in a 2-D cardboard human body model with one additional DOF of width change. We set tip a mixture motion model for body movements and then solve body motion parameters using EM in a statistical framework, where the model-based kinematic constraints are incorporated in a linear form. Tracking results from real video sequences are encouraging.