Research On Key Technologies Of Compressed Sensing For Human Motion Pattern Recognition

Recognition of human motion patterns based on acceleration signals is currently one of the hot topics in the field of pattern recognition.This technology uses the three-axis acceleration sensors built into intelligent terminal devices to sample the acceleration signals of human motion,and then performs feature analysis and pattern recognition.Compressed sensing(CS),as a commonly used signal sampling method,can compress and sample signals at a lower sampling rate,and can recover the original signal from fewer observation data using reconstruction algorithms.However,there is still room for improvement in the reconstruction accuracy and efficiency of compressed sensing.The measurement matrix and reconstruction algorithm are two important steps in implementing compressed sensing,and their optimization directly affects the performance of compressed sensing.Therefore,based on compressed sensing theory,this paper conducts in-depth research and optimization on the measurement matrix and signal reconstruction,aiming to improve the reconstruction effect of acceleration signals and ensure the accuracy of human motion pattern recognition.The main work is as follows:(1)To address the problem of low optimization accuracy in non-diagonal element constraints of the current Gram matrix,an improved measurement matrix optimization method,LBFPhi(L-BFGSPhi),is proposed.This algorithm is used to solve the target matrix function.The L-BFGS(Limited memory-Broyden-Fletcher-Goldfarb-Shanno)method is adopted to determine the search direction,and then the linear search criterion is combined to optimize the search step length and improve the accuracy of the optimal solution of the target optimization function.Finally,based on the acceleration signals of human motion,it is verified by experiments that the optimized measurement matrix can reconstruct the original signal with higher accuracy,which provides better data information for human motion pattern recognition.(2)For the classical fast iterative soft threshold algorithm,an improved signal reconstruction algorithm FIPITA(FISTA-Parameter-Improved Threshold function-Algorithm)is proposed.The algorithm adopts an optimized threshold function,combines the restart mechanism,and introduces three parameters of p,q and r,so that it has higher convergence and better reconstruction performance.Experimental results show that by incorporating the improved reconstruction algorithm into the compressed sensing process,the FIPITA proposed in this paper can achieve both good reconstruction quality and fast reconstruction speed compared with other improved algorithms.(3)This paper proposes a compressed sensing method,called DSDMMV(Double Sparse Denoising and Multiple Measurement Vectors),based on compressed sensing for human motion pattern recognition application.This method addresses the problem of noise interference in data signals during the sampling process and the three-axis multi-channel characteristics of acceleration signals.Furthermore,the LBFPhi measurement matrix optimization algorithm and FIPITA signal reconstruction algorithm are combined with the DSDMMV method and applied to human motion pattern recognition.Comparative experiments with existing research demonstrate that the proposed method achieves higher accuracy in reconstructing human acceleration signals and higher recognition accuracy in subsequent human motion pattern recognition.