Research On Human Gait Recognition Algorithm Based On Deep Learning

The gait feature is non-invasive remote feature and does not require high-resolution images,which has attracted widespread attention and has great potential as security recognition.As an emerging biological feature,gait recognition still has many challenges,such as the recognition of complex background,perspective and clothing change.In order to enable gait recognition technology to be applied in real life as soon as possible,the paper researches gait recognition technology and conducts experiments in the CASIA-B.The main work is as follows:(1)Research on gait recognition based on improved Siamese network.Due to the particularity of gait samples,the paper introduces Siamese network to study gait recognition research,and introduces attention mechanism on the basis of Siamese network.The self-attention model is added to the two-branch network,and the companion loss function is added to the hidden layer to monitor the middle layer during training.After comparative experiments on the CASIA-B,the results show that the average recognition rate is improved by 8.03% and 3.19% in the case of cross-state and cross-view respectively.(2)Research on gait recognition based on Triplet Network.The paper discusses,improves and experiments on Siamese network.The Triplet network is also a network based on metric learning,and the inputs are three images,which are positive,anchor,and negative samples.After the feature extraction network,the distance between the anchor and the positive samples in the embedded space is obtained,and then Triplet loss function is trained to make the distance smaller,and the distance between the anchor and the negative samples larger.The experimental results on the CASIA-B database show that the Triplet network has obvious advantages,and the average recognition rate increasing by 5.42% and 1.23% in the case of cross-state and cross-view respectively.