Markerless Extraction of Gait Features using Haar-like Template for View-Invariant Biometrics

Many research studies have recently shown the possibility of recognizing people by the way they walk i.e. gait. This research is mainly fuelled by the wide range of potential applications where gait biometrics could be useful as the case of visual smart surveillance and forensic systems. In this research paper, we present a Haar-like template for the temporal markerless extraction of gait features under various camera viewpoints. A markerless model-based method whereby angular model templates describing the human motion are employed to guide the extraction process. Gait features consist of the angular measurements for the lower legs in addition to the spatial displacement of the human body. To further refine gait features based on their discriminatory potency, a feature selection algorithm is applied using a newly proposed validation-criterion based on the proximity of neighbours belonging to the same class. Experimental results revealed that gait angular measurements derived from the joint motions can achieve a correct classification rate of 73.6% after applying a rectification process back into the sagittal plane.