A Learning Method for Non-fixed Speed Gait Classification

This paper presented a new human gait classification based on the notion that gait types can be analyzed into a series of consecutive postures types. Silhouettes are extracted using the background subtraction method and then human posture silhouettes features are represented by moment. In the learning stage, we propose a method for the establishment of the standard gait base matrix by the method of recursion. The incoming silhouettes and the database silhouettes are estimated comparison using the Moment Invariants Distance (MID) method. And the comparison result leads to a matrix. In the classification stage, the Minimal Standard Deviation (MSD) algorithm for vector of each motion type is proposed to deduce behavior classification of walker at flexible speed. Finally the method was validated in the outdoor environment an evaluation of eight kinds of gaits involving standing, bending, sitting, walking, jumping, crouching, uphill and downhill is given,  followed by a comparison with other methods. The results show that the identification method presented in this paper has higher rate.