Recognizing wrist pulse waveforms with improved dynamic time warping algorithm

Abstract

This paper proposes an improved DTW algorithm to recognize pulse waveform. Our approach avoids the pathologic alignment that can be caused by original DTW algorithm and DDTW algorithm. Extensive experiments on 1000 pulse waveforms demonstrate the effect of our approach. Compared with original DTW and DDTW algorithms, our algorithm reduces the FRR and FAR greatly. By calculating the similarities between the pulse sample and the five pulse templates, our approach has 92.3% agreement rate with expert.