Parametric elliptic fourier descriptors for automated extraction of gait features for people identification

The interest in gait as a biometric is strongly motivated by the urgent necessity for automated recognition systems for surveillance applications and forensic analysis. Many studies have now shown that it is possible to recognize people by the way they walk i.e. Gait. As yet there has been little formal study of people recognition using the kinematic-related gait features. In this research study, we have investigated the use of Elliptic Fourier Descriptor for the temporal markerless extraction of human joints. We describe a model-based method whereby spatial model templates for the human motion are described in a parameterized form using the Elliptic Fourier Descriptors accounting for the different variations of scale and rotation. Gait features include the angular measurements of the legs as well as the spatial displacement of the body trunk. To further refine gait features based on their discriminability, a feature selection algorithm which is applied using a proposed validation-criterion based on the proximity of neighbors. Initial experiments have revealed that gait angular measurements derived from the joint motions mainly the ankle, knee and hip angles embed most of the discriminatory potency for gait identification.