Skeleton based Frontal Gait Recognition utilizing Fourier Descriptors

Human identification using gait is an emerging field in biometrics. Gait based recognition has a major advantage of being unobtrusive which gives it a cutting edge over other biometric techniques. This paper focusses on gait recognition for narrow corridor and other similar surveillance applications where only the frontal view is available. The skeleton joint information drawn from Kinect is used to obtain the distance between the spine base and each of the joints. This distance function called the centroid distance is used to derive Fourier Descriptors for 5 key frames in a gait period. By using the depth information provided by Kinect, accurate distances between joints is obtained from the frontal view which is advantageous when gait is captured in narrow corridor like places. The extracted Fourier Descriptors are then used along with a kNN classifier for recognition. The proposed method provides considerable increase in recognition accuracy and is also computationally efficient in comparison to existing gait recognition approaches.