Effective Distribution of Local Intensity Gradient Technique for View Invariant Gait Recognition

The proposed paper investigates the effectiveness of view invariance in gait recognition by implementing HOG behavioral feature extraction technique on CASIA-B and CMU MoBo gait database in which standard HOG (RHOG), circular HOG and MHOG are considered for feature extraction. The effectiveness of each feature is analyzed and compared using SVM based classifier on gait detection of the subject for changing view angle. 25 subjects are considered with 10 different view angles for each subject. Classification is done based on the influence of individual and combination of above mentioned features. In spatial domain, although the RHOG gives better precision in finding the gait with normal view angle, but when the view angles are changed with respect to binning angles of histograms, the CHOG feature gives up to 97% better and consistent classification rate against RHOG. Further MHOG feature analysis is considered to improve classification results up to 100%, thus addressing the rotational invariance problem. The work carried out shows better gait recognition results than the previous researchers for all the view angles.