Research On Gait Object Detection And Recognition Method Based On Deep Learning

Gait recognition technology has become the focus of research in the field of biometrics because of its long distance,difficult to disguise and need not be detected.With deep learning gaining prominence in image classification,gait recognition methods based on deep learning have been widely studied due to its simple algorithm structure and strong applicability.However,existing gait recognition methods generally have problems such as target detection accuracy is easily reduced by the changes of multiple covariables(carrying backpack,wearing overcoat),large number of model parameters and gait recognition accuracy is easily decreased by the changes of perspective.Therefore,to solve these problems,this paper studies the gait target detection method and lightweight gait recognition method respectively.The specific research content is as follows:(1)In order to solve the problem that the accuracy of gait target detection is easily reduced due to the change of multiple covariables,a gait target detection method based on Multi-layer Semantic Feature Fusion Network is proposed in this paper.The improved Res Net-18 network was used as the backbone network for target feature detection.The Pyramid Pool Module was used to take into account the local and global information,and the Multi-layer Semantic Fusion structure was designed based on the Feature Pyramid Attention module to realize the retention of multi-level semantic information.Finally,the target was effectively detected and segmented,and the Gait Energy Image was synthesized.Based on the CASIA-B Gait Dataset,the experimental research on the proposed gait target detection method is carried out.The results show that the proposed gait target detection method based on the multi-layer semantic feature fusion network has more accurate and detailed target detection and segmentation effects than other detection methods.(2)In order to solve the problem of the large number of network model parameters and the decrease of the accuracy of gait recognition caused by the change of view angle,a high accuracy gait recognition method based on involution neural network is proposed in this paper.Firstly,this paper proposes the involution backbone network model based on the residual network architecture and involution neural network operator,which uses the involution layer module to extract the gait features to reduce the training parameters of the model.Then,based on the involution backbone network model,a joint loss function composed of strongly constrained Triplet loss function and Softmax loss function is established in this paper,which makes the proposed model have better recognition performance and higher recognition accuracy cross-view conditions.Finally,the experimental study is conducted based on CASIAB Gait Dataset.The experimental results show that the number of parameters of the proposed network model is only 5.04 MB,which is 53.46% less than that of the residual network before improvement.In addition,the proposed network has more lightweight and efficient features,and the proposed method has better recognition accuracy than the mainstream algorithms under the same viewing angle and cross-view conditions.