Study On Technology Of Remote Recognition For Human Activity Pattern From Energy-Efficient Wireless Body Sensor Networks

Recently, the technology of remote recognition for human activity pattern from wireless body sensor networks has been widely attracted attention in remote telemedicine applications, which greatly contributes to the prevention of falls of elderly, cardiovascular chronic disease rehabilitation evaluation. Especially, the above works have been growing to a novel research interesting in some study areas such as the computer science technology, biomedical engineering, and so on. Currently, there still exist some challenging issues including high energy consumption of wireless body sensor networks for activity data acquisition and the poor generalization. Thus, the novel approaches of analysis of activity data is needed urgently for solutions that are found to these issues. In this paper, the advanced compressed sensing theory and sparse representation classification model is introduced for the quantitative analysis of human activity data, in order to develop some novel techniques for the above issues. These works are described as follows:1. A novel compressed sensing framework for activity data from wireless body sensor network with energy-efficient is proposed to reduce energy consumption as much as possible in the WBSN-based system. Its basis idea is that the original activity data acquired from sensor node in WBSN is linearly compressed by the optimal scheme of sparse binary matrices before transmission, and then the block sparse Bayesian learning algorithm is used to improve non-sparse activity signal reconstruction performance, and perfectly reconstruct the original signal. Our proposed method can provide a novel approach for further implementation such as the development of WBSNs-based system with lower energy consumption for remote recognition physical activity.2. A new method for perfect reconstruction of activity signal by using empirical mode decomposition and wavelet denoising is proposed. Its basic idea is that the denoised activity signal is firstly gained by using empirical mode decomposition and wavelet threshold denoising method. And then, considering compressed sensing and activity signal with block structure, block sparse Bayesian learning algorithm is applied to perfectly reconstruction denoised activity signal. Our proposed method can solve the issue of the deterioration of reconstruction performance in compressed sensing for activity signal with poor sparsity that is produced by noise, thus provide reliable data support for further remote activity pattern recognition study.3. A fast sparse representation classification method for human activity recognition based on random projection is proposed, in order to minimize the energy consumption and accurately recognize human activities from WBSNs-based telemonitoring system of human daily activity. The basic idea of the proposed method is that the random projection way of compressed sensing theory is used to reduce the amount of sampling on sensor nodes within WBSN, and then the smaller number of nearest neighbor training sample within the neighbor classed of testing sample, which can optimally linear reconstruction testing samples, are obtained to construct the training sample set of the sparse representation of testing sample. Thus, a fast sparse representation classification algorithm with superior performance of generalization can be developed for capturing valuable features of human activity and improving the recognition rate on the basis of the lower energy consumption and computing complexity of algorithm.4. A joint sparse nearest neighbor representation classification method for human activity recognition based on distributed compressed sensing is proposed, in order to solve the higher computational complexity and lower recognition rate problem of traditional joint sparse representation classification algorithm, which ignoring the characteristics of temporal and spatial correlation between collected data from multi-sensor nodes. Its basic idea is that the distributed random projection way of distributed compressed sensing is used to reduce the transmission energy consumption of WBSNs-based sensor nodes, and then a smaller-optimized overcomplete dictionary set, which can optimally linear representation test samples, is obtained to construct the joint sparse nearest neighbor representation mode based on activity compressed data of multi-sensor nodes. The proposed method can effectively reduce the computational complexity of algorithm, and improve the multi-sensor nodes-based human activity pattern recognition accuracy rate.The activity data from opened single-sensor-based USC-HAD dataset of American University of Southern California, and multi-sensor-based WARD dataset of University of California at Berkeley are used to objectively evaluate the effectiveness of our proposed methods, in which provide new ideas and methods for the design of energy-efficient WBSNs-based human activity telemonitoring system.