Front-view Gait Recognition

We present a new method for front-view gait biometrics which uses a single non-calibrated camera and extracts unique signatures from descriptors of a silhouette's deformation. The proposed approach is particularly suitable for identification by gait in the real world, where the advantages of completely unobtrusiveness, remoteness and covertness of the biometric system preclude the availability of camera information and where the CCTV images usually present subjects from an upper front-view. Tests on three different gait databases with subjects walking towards the camera have been performed. The obtained results, with mean CCR of 96.3%, show that gait recognition of individuals observed the front can be achieved without any knowledge of camera parameters. Moreover, the method has been applied to three different walking directions and the results have been compared with the algorithms found in literature. The performance of the proposed system is particularly encouraging for its appliance in surveillance scenarios.