Multi-attributes gait identification by convolutional neural networks

Gait as a biometric feature that can be measured remotely without physical contact and proximal sensing has attract significant attention. This paper proposes to use con-volutional neural networks (ConvNets) and multi-task learning model(MLT) to identify human gait and to predict multiple human attributes simultaneously. In comparison to previous approaches, two novelty in our convolutional approach can be summarised as (i)using ConvNets to learn rich features from the training set is more generic and requires minimal domain knowledge of the problem compared to hand-craft feature, (ii) to identify human gait and to predict other human attributes simultaneously can achieve improved performance for all task than standalone gait identification. Specifically, we first extract Gait Energy Image(GEI) from each walking period as the low level input for the ConvNets. Secondly, we train the ConvNets through back-propagation using a joint loss of each task. Finally, high-level feature is hierarchically extracted in ConvNets, which is shared by each task and used to identify human gait and to predict attribute. The approach was verified on CASIA gait database B, achieving over 95.88% accuracy for each task. To the authors' best knowledge, this is the first time multi-attributes gait identification being proposed.