Non-local gait feature extraction and human identification
- PDF / 1,163,234 Bytes
- 14 Pages / 439.642 x 666.49 pts Page_size
- 67 Downloads / 193 Views
Non-local gait feature extraction and human identification Xiuhui Wang1
· Wei Qi Yan2
Received: 22 December 2019 / Revised: 15 September 2020 / Accepted: 17 September 2020 / © Springer Science+Business Media, LLC, part of Springer Nature 2020
Abstract As a new human identification technology, gait recognition is receiving more and more attention in recent years. However, traditional gait recognition techniques are limited by the challenges of feature representation and extraction algorithms. In this paper, by utilizing the self-attention mechanism, we propose a novel gait-based human identification solution. Firstly, we utilize non-local neural networks (NLNN) to extract non-local features from a pair of randomly selected gait energy maps (GEIs). Secondly, based on the relationship between GEIs and various parts of the human body, the output of NLNN is horizontally segmented into three sections, i.e., strong-dynamic region, weak-dynamic region and microdynamic region, respectively. Thirdly, the segmented gait features are weighted ensembled by three two-class classifiers. Finally, two experiments are carried out with the OU-ISIR large population dataset and the CASIA dataset B to evaluate the proposed approach. Keywords Human identification · Non-local features · Gait recognition · Self-attention
1 Introduction As a prominent human identification technology, gait recognition is receiving more and more attention in recent years [9, 26]. Gait recognition can be used in intelligent video surveillance systems and has also been investigated as new means in human leg rehabilitation medical diagnosis. Compared with other biometric methods, such as face recognition and fingerprint recognition, gait recognition has the advantage of easy remote recognition, and the recognition process usually does not need to be deliberately cooperated by the recognized object [33]. However, due to changes in external factors during gait data collection, Xiuhui Wang
[email protected] Wei Qi Yan [email protected] 1
Key Laboratory of Electromagnetic Wave Information Technology and Metrology of Zhejiang Province, College of Information Engineering, China Jiliang University, No. 258, Xueyuan Street, Hangzhou 310018 China
2
Auckland University of Technology, No. 2-14, Wakefiled Street, Auckland 1010, New Zealand
Multimedia Tools and Applications
such as lighting, road conditions, camera resolution as well as clothing, weight-bearing, and carrying conditions of a pedestrian, gait variance of the same person may be much obvious than the differences from different persons. The traditional method to resolve the problems is to construct well-designed gait features and reduce the influence of interference on gait recognition by setting a group of constraints [17, 22, 23]. Nevertheless, though this kind of methods can well solve the problem in a specific environment, it is difficult to apply them to other applications. Fortunately, deep learning techniques provide better support for gait feature representation using end-to-end technology [
Data Loading...
