*3.1. Identification Algorithm Valid Segmentation Results*

For the 65 groups of motions with the same label, we calculate the Jaccard index of each channel of acc and ypr and then determine the average Jaccard index for each motion by averaging the six-channel values. As shown in Table 2, all the extracted valid segments' indices except lying on the desktop and writing notes are more than 88% similar to the benchmark. The Jaccard index of lying on the desktop and writing notes is worse than other motions, which may be due to the sensor data not changing significantly during motion times, as well as the warped path between the adjacent paths being near. This is a weakness in our proposed VB-DTW algorithm, which makes the algorithm inefficient for long-term recognition of a substantial portion of near-static data. We will continue to investigate the most effective approach to dealing with precise and effective segment extraction in subsequent tests.

**Table 2.** Jaccard index for 13 motions. All the extracted valid segments' indices except lying on the desktop and writing notes are more than 88% similar to the benchmark.

