Analysis of filtering methods for 3D acceleration signals in body sensor network

Development of denoising algorithm for 3D acceleration signals is essential to facilitate accurate assessment of human movement in body sensor networks (BSN). In this study, firstly 3D acceleration signals were captured by self-developed nine-axis wireless BSN platform during 12 subjects performing regular walking. Then, acceleration noise was filtered using four common filters respectively: median filter, Butterworth low-pass filter, discrete wavelet package shrinkage and Kalman filter. Finally, signal-to-noise ratio (SNR) and correlation coefficient(R) between filtered signal and reference signal were determined. We found that (1) Kalman filter showed the largest SNR and R values, followed by median filter, discrete wavelet package shrinkage and finally Butterworth low-pass filter; whereas, after correcting waveform delay for Butterworth low-pass filter, its performance was a little better than that of Kalman filter; (2) Real-time performance of median filter related to its window length; Decomposition level influenced real-time performance of discrete wavelet package shrinkage; Butterworth low-pass filter could bring large waveform delay if filter order and cut-off frequency were not properly selected. The algorithms of these filters would be further investigated to achieve best noise reduction of 3D acceleration signals in future.