An application for gait recognition using persistent homology

Abstract

This Demo presents an application for gait recognition using persistent homology. Using a background subtraction approach, a silhouette sequence is obtained from a camera in a controlled environment. A border simplicial complex is built stacking silhouettes aligned by their gravity center. A multifiltration is applied on the border simplicial complex which captures relations among the parts of the human body when walking. Finally, the topological gait signature is extracted from the persistence barcode according to each filtration. The measure cosine is used to give a similarity value between topological signatures. The input of this Demo are videos with resolution 320x240 to 25f ps. The videos in CASIA-B database are used to prove the efficacy and efficiency. A computer with 2Gb of RAM memory and a DualCore processor was used to test the implementation of the proposed algorithm. In this Demo all related tasks have been programmed by the authors in the C++ programming language. OpenCV library has been used for the image processing part.