L1-Norm Distance Metric Learning for Gait Recognition

Gait recognition is a biometric technology which realizes human identification through the characteristics of human movement during walking. It has a wide range of potential applications in security, surveillance, medical and other fields. In this paper, we propose a L1-norm distance Metric Learning (L1 ML) method to study the problem of gait recognition. The proposed L1 ML aims to learn a linear transformation under large margin framework so that the L1-norm distance between positive sample pairs in the transformed subspace is smaller than a small threshold, while that of negative sample pairs in the transformed subspace is greater than a large threshold, which can effectively distinguish samples of different subjects. Unlike traditional methods that employ L2-norm to calculate the distance between samples, our L1ML utilizes L1-norm distance, which is more robust to outlier samples. We conduct a series of comparative experiments on the CASIA-B gait dataset, and the experimental results verify the effectiveness of the proposed method.