Research On Gait Cycle Detection Method

Gait recognition technology is receiving more and more attention with the development of biometrics,and most gait recognition methods are based on a good period detection.Based on the traditional periodic detection method,this paper analyzes the advantages and disadvantages of the traditional periodic detection methods and presents an evaluation index for the performance of periodic detection accuracy.In this paper,according to the problems existing in traditional gait cycle detection and the characteristics of gait database,a deep convolution neural network is used to detect gait cycle.The content of this paper is developed from two directions.One is to use periodic classification which is usually used by convolutional neural networks to carry out periodic detection;the other is to carry out periodic prediction by using the convolutional neural network framework to carry out regression prediction by making changes.The idea of classification is to classify the images of the original database according to a fixed category,label them,and then input the marked database into the network for training.According to the difference in the number of classification categories,experiments are conducted to obtain each network.The recognition rate of the model and its classification number is obtained by inputting an unmarked contour sequence into the trained network model to obtain the classification of each frame of the contour sequence.The gait cycle under the sequence is obtained according to the classification result.The idea of regression is obtained by further optimization methods after considering the problems of classification.During the experiment of classification,we found the problem of misclassification between the adjacent frames in the classification.In response to this problem,we consider using the regression method for periodic detection.The regression method changes the output layer of the original network frame and outputs the activation function to output it as a floating point number corresponding to the frame image.The network parameter is adjusted through constant comparison with the value of the marked sample to achieve the purpose of training network.Based on the realization and analysis of the traditional periodic detection methods,this paper presents the methods of using convolutional convolution neural network classification and convolutional neural network regression.On the basis of the trained model,an untagged gait profile sequence is output in the respective network model,and the number of frames in one cycle of the sequence is obtained.By comparing with the traditional method,we find that we can use the deep convolution neural network to detect the gait cycle.Based on the classification and regression-based methods,we can get good results in multiple perspectives.Experiments show that the method of deep convolution neural network can solve the problem of feature extraction well in the front and back view and the period of the sequence can be well approximated to the standard period by using this feature.The traditional method has a strong advantage.The method of deep convolution neural network can ensure that the extraction cycle will not be greatly deviated.In general,the deep convolutional neural network can ensure that the error between the number of periodic frames in the side view to the standard number of frames is small,and the number of frames extracted for the front and back view angles can have larger improve compared to the traditional method.