Successive Human Tracking and Posture Estimation with Multiple Omnidirectional Cameras

We propose a successive method for human tracking and posture estimation by using multiple omnidirectional cameras appropriate for Machine Learning method. A stable estimation for foot and head position is executed by the combination analysis with particle filter processing. Moreover, a classification method is accomplished by using the constraint of the connected line between head and foot position. The combination both this constraint and relative height from head to foot is possible to distinguish typical four postures for human activities in an indoor scene. We believe that this continuity of each data helps smooth convergence to the time-sequential learning for the discrimination between normal and abnormal behavior.