Research And Implementation Of Gait Recognition System In Complex Scene

Gait is recognized as a suitable feature for long distance person identification.However,gait is also affected by wearing and perspective,which results in poor gait silhouette extraction and low cross-view recognition rate in complex scene.In view of these shortcomings,this paper proposes a gait silhouette extraction model for complex background,and proposes some gait recognition methods based on gait energy image and gait sequence to improve the robustness of gait recognition in complex scene.The main research contents include the following three aspects:(1)Based on gait energy image,a new method of gait recognition is proposed,which combines nonlocal and part features.Using the non-local neural network and human horizontal partition,combined with the nonlocal and part features of gait,to enhance the ability of classification.(2)In order to solve the lack of time-domain information in gait energy image,silhouette sequence is used as gait feature.Firstly,a traditional gait recognition method based on adaptive hidden Markov model is proposed,and temporal features are extracted by temporal modeling.Secondly,a sequence based multi-scale gait recognition method is proposed.Multi-branch convolutional network is used to learn the part features of different scales separately,and attention based pooling algorithm is used to reduce the impact of some poor quality silhouette on the overall features.(3)A lightweight semantic segmentation network for gait silhouette extraction in complex background is proposed.The residual network ResNet-18 is used to quickly encode high-level semantic information,and an additional branch is used to supplement the lack of spatial information.The network adds contour auxiliary loss to improve the accuracy of segmentation,and finally a simple gait recognition system is built.Through the above research and the experimental results in two large datasets CASIA-B and OU-ISIR-LP,our approach can be proved effective.It solves the difficulty of gait silhouette extraction in complex background,enhances the ability of gait feature extraction and effectively improves the robustness of gait recognition in complex scene.