Performance analysis on new biometric gait motion model

Recognising people by the way they walk and/or run is new. A novel analytical model which is invariant to human gait of walking and running is developed based on the concept of dynamically coupled oscillators and the biomechanics of human walking and running. It serves as the foundation of this automatic person recognition system. The effects of noise and low resolution have been evaluated on the largest data set of its kind. This is useful as security camera footage is usually prone to noise and of poor resolution. The gait signature is formed from the Fourier description of the thigh and lower leg rotation. Angles of rotation are extracted via temporal template matching across the whole image sequence. Classification is done via the k-nearest neighbour and cross-validated with the leave-one-out rule. The promising recognition rates for both walking and running suggest the high potential of this technique and using gait as the cite for person identification in practical applications. Future work will focus on understanding the features used to create the gait signature in order to further improve the recognition rate and will determine the invariance attributes for walking and running.