Memory-based Gait Recognition

Dan Liu, Mao Ye, Xudong Li, Feng Zhang and Lan Lin

Abstract

Gait recognition is an interesting and challenging task aiming to classify the subjects based on the way they walk, which is subject to various covariates including carrying, clothing, surface and view angle. In this paper, we propose to utilize the memory mechanism to effectively alleviate the aforementioned problems. Specifically, we extract the 2D location information of human joints as the gait features via the migratory articulated human detection. Inspired by the mechanism of brain sequence processing, we input the gait feature sequence into the memory-based gait recognition (MGR) network, which achieves the process of memory and identification of the gait sequence. Our proposed MGR is robust to the noise that maybe exist in the gait sequence features. Besides, MGR is able to learn on the data with long range temporal dependencies. The experimental results on the CASIA A and CASIA B gait datasets verify the feasibility and effectiveness of the proposed method.

Session

Posters 2

Files

Extended Abstract (PDF, 334K)
Paper (PDF, 3M)

DOI

10.5244/C.30.82
https://dx.doi.org/10.5244/C.30.82

Citation

Bibtex

        @inproceedings{BMVC2016_82,
        	title={Memory-based Gait Recognition},
        	author={Dan Liu, Mao Ye, Xudong Li, Feng Zhang and Lan Lin},
        	year={2016},
        	month={September},
        	pages={82.1-82.12},
        	articleno={82},
        	numpages={12},
        	booktitle={Proceedings of the British Machine Vision Conference (BMVC)},
        	publisher={BMVA Press},
        	editor={Richard C. Wilson, Edwin R. Hancock and William A. P. Smith},
        	doi={10.5244/C.30.82},
        	isbn={1-901725-59-6},
        	url={https://dx.doi.org/10.5244/C.30.82}
        }