Gait Recognition And Simulation Based On The Ensemble Deep Learning

Gait recognition and simulation are the research hotspots in the field of biometrics and computer vision.The Gaussian-based Conditional Restricted Boltzmann Machine(GCRBM)time series model could efficiently predict a single type of time series data,however,it could hardly identify or predict multi-category time series data,and it has not been used for the recognition and simulation of actual gait data.In hence,in this paper we utilize the excellent performance of deep learning and combines with the ensemble learning techniques to improve gait recognition accuracy while achieving gait movement simulation,two types of Gait Recognition and Simulation algorithms based on the Ensemble Deep Learning are proposed.The innovation of this paper is described as follows:1.The gait recognition and simulation algorithm based on the ensemble deep belief networks(DBN)is proposed.As to the problem that the GCRBM model was difficult to identify and predict actual multi-category gait time series data,the proposed model integrates multiple deep-network combinations with DBN and GCRBM,the DBN model in each deep-network learns the low-dimensional features of each gait time series and the low-dimensional feature will be used to train the GCRBM model.In the step of gait recognition and simulation,the ensemble DBN is used to identify the class of gait data by using the minimum reconstruction strategy,and then the latter gait time series will be predicted by the corresponding GCRBM model.Finally,the gait images can be reconstructed by the corresponding DBN model.The experimental results on CASIA gait datasets of CAS shows that average gait recognition rate of the proposed model can be improved 10% ~ 20% when compared to the relevant methods,and it can predict and simulate the latter gait time series.2.The gait recognition and simulation algorithm based on the ensemble Convolutional Neural Network(CNN)and deep belief networks(DBN)is proposed.As to the problem that the recognition performance of integrated DBN is not good enough,the CNN model is introduced into the gait recognition stage.The proposed model uses all kinds of gait data to train several different structure of CNN-Based Classifier and then to train the integration of DBN and GCRBM.In the step of recognition and simulation,the proposed model will identify the class of gait with all CNN-based classifiers by using the "minority-obeying" voting strategy,and then the latter gait time series will be predicted by the corresponding GCRBM model.Finally,the gait images can be reconstructed by the corresponding DBN model.Similarly,the experimental results on CASIA gait datasets of CAS shows that the average gait recognition rate of the proposed model can be improved 5% ~ 8% when compared to the method of integrated DBN,and it also can predict and simulate the latter gait time series.3.A gait recognition and simulation system based on the Ensemble Deep Learning is built.We can conduct the experiments of gait recognition and simulation and verify the effectiveness of all proposed algorithms through the System.