View-Invariant Gait Recognition With Attentive Recurrent Learning of Partial Representations

Gait recognition refers to the identification of individuals based on features acquired from their body movement during walking. Despite recent advances in gait with deep learning, variations data acquisition and appearance, namely camera angles, subject pose, occlusions, clothing, are challenging factors that need be considered for achieving accurate systems. In this article, we propose a network first learns extract convolutional energy maps (GCEM) frame-level features. It then adopts bidirectional recurrent neural learn split bins GCEM, thus exploiting relations between learned partial spatiotemporal representations. We use an attention mechanism selectively focus important recurrently representations as identity information different scenarios may lie GCEM bins. Our proposed model has been extensively tested two large-scale CASIA-B OU-MVLP datasets using four test protocols compared number state-of-the-art baseline solutions. Additionally, comprehensive experiment performed study robustness our presence six synthesized occlusions. The experimental results show superiority method, outperforming state-of-the-art, especially where clothing carrying conditions encountered. also revealed is more robust against occlusions methods.