Gait features obtained by current extraction methods are easily affected by people's walking direction, dresses, and carryings, due to which gait recognition system has not yet appeared. An extraction method based on centroid is proposed in this chapter. Segment and track the moving silhouettes of a walking figure in image sequences to calculate the silhouettes' centroid. The complex silhouette is represented by a point to avoid the influence of dresses and carryings. Divide centroid coordinate value by the height of detecting walking figure to normalize to remove the disturbance caused by walking direction relative to the camera optical axis angle. By denoizing centroid trajectory remove the noise caused by some accidental factors to obtain regular wavelet curve whose main frequency component distribution vector is the final gait feature. Experimental results show that this approach can obtain identical gait features even when experimenters change their walking directions, dresses, or carryings, tolerating noise and low resolution.
展开▼
