A multiscale morphological method for human posture recognition

For the purpose of estimating the posture parameters of moving human bodies in visual surveillance applications, we present a model-based shape analysis method. The problem is converted to the optimal matching of the 2D silhouette of the human body in parametric shape space. We point out the causality of the area-difference-based shape similarity in morphological scale space, which contributes a lot to the robustness and reliability of the matching. Based on this causality, the method we propose works in a course-to-fine manner on cracked silhouettes practically segmented from complex environments, and can converge fast enough to meet real-time requirements.