Research On Multi-view Gait Recognition Based On Deep Learning

Gait recognition is an emerging biometric recognition technology.Compared with traditional biometric recognition technology,it has the advantages of non-contact,longdistance recognition,and difficult to disguise.It has been widely concerned about research.However,there are many influencing factors in the gait recognition process,including angles,pedestrian dressing,etc.,which will have a great influence on the recognition accuracy.Deep learning methods has obvious advantages in image recognition and feature extraction.This paper has studied gait recognition methods based on deep learning.The specific research contents are as follows:A multi-view gait recognition model based on generative adversarial networks is proposed.Perspective conversion is an effective method to resolve gait recognition,that is,to convert the gait image of any perspective to a specific perspective.However,the conversion of a specific perspective of the gait image will cause the loss of the gait features.In order to solve the loss of features in the process of perspective conversion,multiple models are set up to convert multiple perspectives simultaneously,and multiple generated gait images are used to retain different feature information.Generative adversarial networks is used to build a multi-view conversion model in view of its excellent performance in the field of image processing.This makes it possible to use more feature information to determine the identity of pedestrians during the recognition process and achieve the goal of improving the recognition rate.In the proposed model,both the number of angles and the view angles will affect the accuracy of gait recognition.In order to discover the impact of these factors on the recognition results,the focus is on the number of angles and angle combinations of the model.The corresponding selection strategies are given respectively,and the model parameters considering the recognition rate and efficiency are determined through experiments.Finally,the effectiveness of model is proved by comparison with other existing methodsThe gait recognition method in the previous study uses KNN algorithm to realize pedestrian recognition,but in this paper's multi-perspective transformation model,the KNN algorithm will obtain different neighbors at multiple angles.In order to synthesize these neighbor data and obtain better recognition results,this paper studies the correlation calculation of the KNN algorithm and the weight of the neighbor data,analyzes the influence of different similarity functions on the accuracy of gait recognition,and gives a weighted classification algorithm that considers the weight of the nearest neighbor.Compared with the original method,the results show that choosing the appropriate similarity function can help to improve the recognition rate under the general circumstances of ignoring pedestrian backpacks,wearing and other influencing factors,and considering the weight can comprehensively improve the accuracy of gait recognition under the complex situation of comprehensively considering these influencing factors.