Gait Recognition Based On Convolutional Neural Network

Gait feature is one of the most promising biometric features.Compared with other biometric features,such as human face,fingerprint,etc,gait has the prominent advantages of non-contactness and long distance allowance.With the rapid development of information technology,identifying a target person through gait is of great significance in intelligent video monitoring and identity recognition.In traditional gait recognition methods,the recognition rate is deeply affected by gait models,because the parameters of gait models are selected according to the human experience.In addition,external factors such as coats and backpacks will have great impacts on the training of gait models.To solve the above problems,this paper carried out the following researches:(1)Based on the power capacity of convolutional neural networks(CNN)in feature extraction and representation,we constructed several CNN models for gait recognition.These models utilize convolutional operations to extract gait features from the input gait videos or image sequences,and then use pooling operations to achieve feature selection and information filtering.Finally,a fully connected layer is used to integrate gait features that have been highly abstracted after multiple convolutions.(2)Based on the generalization ability of SVM multi-classifier,we choose SVM as the final gait classifier,in which the original features are mapped into high-dimensional space by selected kernel functions.(3)By utilizing the advantages of Euclidean distance in similar metrics,we integrated the Euclidean distance into the loss layer of our gait models,and refined these models by calculating the minimum distance difference between the real value and the predicted value.