A New Attempt to Gait-based Human Identification

Vision-based human identification at a distance in surveillance systems has attracted more attention recently, and its current focus is on face, gait or activity-specific recognition. Based on principal component analysis, this paper makes a simple but efficient attempt to gait recognition. For each gait sequence, an improved background subtraction procedure is first used to accurately extract spatial silhouettes of a walking figure from the background; Then, eigenspace transformation to time-varying silhouette shapes is performed to realize feature extraction; Gait recognition using spatio-temporal correlation or the nearest neighbor classifier is finally accomplished in lower-dimensional eigenspace, and some additional personalized physical properties are selected for the validation of final decision. Experimental results on a small database show that our algorithm has an encouraging recognition rate with relatively lower computational cost.