A Kinematic Model for Human Motion and Gait Analysis

This work introduces a new model-based approach towards the 3-D tracking of human motion. We suggest a parametric body model, with both coarse and fine resolution levels, characterized by hard and soft kinematic constraints. The hard constraints place physical limitations on possible model configurations while the soft constraints take the form of a priori, stochastic distributions learned from previous examples of human body configurations. This prior knowledge is used to define a natural acceleration of body parameters towards their expected values. Coupled with the hierarchical structural model, these kinematic constraints yield tracking that is robust to complex motion and occlusions. We demonstrate the utility of the proposed model on video sequences of human motion captured in a home environment.