Gait recognition based on capsule network

Abstract

Gait as a biometric feature is widely used for human identification, and gait recognition has recently become a significant research problem. According to a small amount of labeled multi-view, multi-walking-condition and multi-clothes-condition human walking videos, we can find an effective model based on capsule network to capture more discriminative features and promote gait recognition performance. This paper works on gait recognition based on capsule network and we consider two different architectures, namely matching local features at the bottom layer based on capsule network and matching mid-level features at the middle layer based on capsule network, input images such as GEI, CGI, and resolution of input image. Empirical evaluations are conducted in the aspect of kinds scenarios, namely cross-walking-condition, cross-view and cross-clothes condition. The approaches are evaluated on the CASIA-B dataset and OU-ISIR Treadmill dataset B. These results show that the methods exceed the previous state-of-the-art outcomes.