Gait Analysis Based On Deep Neural Network And Deployment In Edge Intelligent Terminal

According to data compiled by the China Academy of Information and Communications Technology,the global biometric identification technology market reached about $20 billion in2019,an increase of $10 billion compared with 2015,which is in a state of rapid development.Moreover,the fingerprint recognition market accounted for 58% of the total market,and the face recognition market accounted for 18%.The iris recognition market accounted for about 7% of the total market,and other recognition technologies such as palm print,voice,and gait accounted for the remaining 17% market share.Due to the unstable performance and high error rate of current gait recognition algorithms,gait recognition is seldom used in real life.Therefore,this paper mainly studies the gait recognition algorithm based on a deep neural network to improve the recognition performance,and designs a gait recognition system to contribute to the development of gait recognition.Firstly,a gait recognition algorithm that combines attention mechanism and residual block with multiple feature extraction blocks in parallel is proposed,and three gait datasets by collecting gait data of 113 people from Nanjing University of Posts and Telecommunications are created.The data processing part of the datasets mainly includes cycle segmentation and data enhancement,and then feature extraction and gait recognition are performed through the proposed model.After the experiment,the best recognition accuracy of the proposed model in the NJUPT II and NJUPT III gait data sets reached 93.37% and 98.56% respectively.Compared with the CNN+LSTM model,the accuracy of the proposed model is improved by 5.07% and 1.24%,which proves that it can effectively improve the recognition effect.Secondly,a compression algorithm is proposed to compress the gait recognition model to simplify the gait recognition model and speed up the inference,in view of the model needs to be deployed on the edge of the resource-constrained intelligent terminal.In this paper,the final pruning scheme is confirmed mainly from the three aspects of model network parameter importance confirmation,pruning threshold setting and model performance recovery to achieve the effect of balancing model performance and size.Experimental results show that the proposed algorithm can achieve a compression ratio of 3 × without sacrificing the model performance,and shorten the inference time of the model for a single sample from 45.85 ms to 14.95 ms,which effectively simplifies and accelerates the model and reduces the difficulty of model deployment in edge terminals.Finally,a gait recognition system is designed and implemented in this paper,which mainly includes information acquisition and identity reasoning modules.In the information collection module,a new gait data set is created by user input personal information and gait signal collection.In the identity reasoning module,the designed model is deployed on an intelligent terminal.Furthermore,it can collect real-time gait data and feed it into the model for inference,and then display the recognition results.The final results show that the system can collect the gait data and real-time identification and then display the identification results.