Research On Human Infrared Images Gait Recognition Method Based On Deep Learning

Human gait is the same as human fingerprints and human iris,which are unique biological properties of the human body.In recent years,many scientific researchers have also invested in the recognition of human gait.However,current related studies are based on studies of human visible light gait that limited general visible light imaging skills in conditions such as heavy rain,snow,fog,and dark nights.The range of the gait recognition is limited.Infrared images can effectively compensate for defects in visible light.Therefore,this thesis extends the relevant gait detection methods of infrared images of human gait.In this thesis,the recognition of infrared human gait image is mainly studied from the following six points:1.Configure the infrared dataset for human gait,first use the binary filtering method to remove the noise in the infrared image and then use the CLAHE method to enhance the image after binary filtering to resolve the contour blur.Use the background subtraction method.It extracts the silhouettes of the human body,removes duplicate spatial information and finally connects the silhouette features such as hair lines and backpacker contours.After preprocessing the above human body infrared image,we will create a human body infrared gait data set that meets the training requirements of the model proposed to this document.2.According to the characteristics of infrared human gait images,single-branched gait recognition model and multi-branched gait recognition model were designed respectively based on convolutional neural network.In order to obtain gait characteristics of different sizes,convolution kernels of different sizes were set in the multi-branch model to obtain receptive fields of different sizes.To build two convolution neural network model,respectively on the self-built infrared human gait data set for a 1000 rounds of iterative training,the experiment tests have shown that many branches of convolution neural network model has better effect on infrared human gait recognition,in which the accuracy of close to 100% in the training set,the accuracy of more than 99.3% on the test set.In addition,the verification set,which is completely disjoint with the training set and test set data,is used for verification,and the recall rate,precision and F1 value of the two network models are compared and analyzed.It is concluded that the multi-branch convolutional neural network is obviously better than the single-branch convolutional neural network in recognition effect.Therefore,the multi-branch convolutional neural network model is finally determined as the final infrared human gait recognition model in this subject.3.In order to verify the necessity of image preprocessing,such as eliminating the redundant spatial information in the original infrared human gait image and bridging the additional contour features,this thesis conducted correlation analysis on the attention of the model.The attention of the model was analyzed by using the original infrared human gait image,the image after eliminating the redundant spatial information,the image after eliminating the redundant spatial information and the image after filling the extra contour features.This experiment can intuitively show the region that the two models are most interested in the human gait contour in the recognition task.4.Based on the multi-branch neural network model built in this thesis,we designed infrared image and human body gait detection system prototype,the user uploaded infrared image of human body gait.human body can be effectively recognized and stored in the database.The detailed identification information about the volunteer is requested and the result of the identification is returned to the front-end and sent.The goal of practical application has been achieved initially.This thesis combines deep learning theory with gait recognition technology to solve the problem of human identity gait recognition in infrared images,which can further expand the scope of infrared technology,which is important for public forensics,video surveillance and human identity recognition in specific situations.