A statistical approach to gait recognition and verification by using cyclograms

The cyclogram of human gait uncovers the relationship between thigh and knee motion. Because it is believed to be invariant for the walking pattern of the same person it leads to a better understanding of gait. If we use it as the signature in gait recognition and verification, it could lead to an automatic person recognition system using video footage from security cameras. The body contours are extracted by image subtraction and edge detection and rotation angles of thigh and lower leg are derived by regression analysis. Then we apply a weighted moving average and a low-pass filter to smooth the angle data. Cyclograms are generated as the thigh angle vs. knee angle and are fitted by non-uniform B-spline curves. To compare the signatures between two gaits, the differences of shape and phase of the cyclogram are calculated using the point projection method and extreme points of curves. Finally, classification is done via the k-nearest classifier and cross-validation with the leave-one-out rule.