Dynamic time warping for IMU based activity detection

Based on an IMU motion sensitive device, a dynamic time warping (DTW) to detect risky activities of people under healthcare has been devised and implemented. DTW, beneficial by matching time sequences with nonlinear dynamic alignment of the sample points to optimize the similarity, is especially a candidate for such a kind of activity detection because that the discrete activity signal sequence would be investigated with various time or speed, and often be unequally lengthened. DTW shows potentially an excellence in dealing straightforward with this kind of unequal-length sequence matching problem. In this study, we clarify the use of DTW for such an application in both theory and application aspects. Besides, a related neural network mapping development, often used in solving the same kind of problem, is intentionally introduced in the paper for comparison. A performance investigation in both developments arise to seek an excellent methodology to deal effctively and efficiently with the activity detection. The comparison causally confirms the direct template matching based method DTW is superior in detecting temporally a matched activity from a continuous signal sequence source. DTW has thus been recommended to incorporate with the electronically assistive IMU device for monitoring specifically risky motions of the people who are under remote healthcare.