Significance Of Effective Local Gradient Distribution Technique With Eft On Multi-View And Cloth Invariant Gait Recognition

Abstract

The gait recognition is gaining a growing interest from computer vision researchers as a trusted second
generation biometrics. Change in walking speed of the subject, variation in cloths and variation in view
angle of the gait poses serious challenges to gait biometrics. The current research has not yet addressed
these challenges in gait recognition. This research work establishes the significance of spatial and
frequency domain feature extraction on cloth and view invariant gait recognition. It considers the feature
extraction algorithms CHOG and EFT and their individual output response to the SVM classifier is
analyzed for cloth and view invariant gait recognition. The circular histogram of oriented gradient
captures the decimated angular local energy gradient feature, the Elliptical Fourier Transform is applied
on CHOG feature gives the accurate geometry of the gait. The fusion of this two-feature extraction
algorithm implicitly captures geometrical structure and dynamically changing characteristics of gait
under different view angle and with different cloths. The CHOG feature effectively contributes for the gait
classification with different view angle but showcases weak features with changing cloths resulting in a
poor classification rate. The combination of CHOG features with EFT gives an excellent geometric
structural details of inter variation gait pattern and has given the classification rate via SVM of 97% and
above for 10 different view angle and 11 different cloths. The combination of CHOG and EFT gives very
encouraging and stable results versus the previously proposed spatial domain techniques like regression
model, entropy feature, deterministic learning, GEI with static and dynamic approach.