Horizontal slicing-based frontal view gait recognition under occlusion with Augmented Reality Annotations

In line with the developments in biometric securities, there is a greater focus to leverage gait as a means of recognition. Gait signifies the unique walking pattern in human beings; it has delivered well at tasks pertaining to person re-identification. Being unique as distinct from other biometric features, gait is unique since it is basically a subconscious behavior, which performs minimization of the risk of purposeful obfuscation due to malicious intentions of subjects. A practical problem in gait recognition, the case where a portion of the input is occluded from the camera-view, is not much researched for Red-Green-Blue (RGB) video modality. It furnishes viable means of performing gait recognition due to recent increase in adoption of Closed Circuit Television (CCTV) cameras for security. In this research, we propose a robust model named GROS-AR leveraging a horizontal slicing mechanism, which shows that extracting features part-wise can help successfully capture gait signature in occlusion scenarios; and after recognition, we can use the same for displaying the pertaining Augmented Reality Digital Annotations. We demonstrate our approach on the publicly available two data-sets Frontal-View Gait (FVG) dataset and CASIA B, the CASIA gait recognition dataset B, and observe encouraging results. Further, we develop an Augmented Reality system that is capable in creating visual descriptions of the identified subjects on top of the video frames to visualize person identification.