Gait Recognition using Time of Flight Sensor

Abstract

Motion capture and analysis has an important role in many fields, therefore it has been for long time and still is an active research field. A large number of applications can benefit from motion capture and analysis, including athlete training, medical diagnostics, surveillance, and video retrieval. Conventional motion capture relies on attaching sensors or markers to the subject’s body. While most recent motion capture approaches are markerless and try to extract biomechanical models and features, based on non-intrusive techniques such as video based approach. This project aims to develop a video analysis system for extracting gait features based on depth and intensity frames acquired by Time-of-Flight (ToF) camera . The main steps of the project are as follows: Segmentation to extract body silhouettes from frames using depth information provided by the 3D ToF camera which represents the distance from the object to the camera for every pixel in the intensity frames. After that we apply morphological filters to enhance the segmented object and eliminate the background noise. The enhanced silhouette is then divided into body segments based on human anatomical knowledge. Then we apply ellipse fitting techniques on the body segments to obtain a set of topological features of the silhouette. The resulting gait pattern representing the evolution of these topological parameters in time is called the gait pattern or signature and used to characterize the gait of a person. Once these signatures are obtained for every subject in the test group, we conduct several tests of robustness of their signatures for identification. The results obtained so far show very good recognition rates. This master thesis has resulted in one conference paper in BIOSIG 2011.