| Summary: | 碩士 === 元智大學 === 工業工程與管理學系 === 105 === Wearable inertial sensors are used in a large number of motion sensing, for the measured data to restore the subjects action. But the current sales of mobile bracelet action measurement, can only be the entire data analysis of sports performance, can not be targeted for specific action analysis and processing. Therefore, this study uses the dynamic time correction algorithm (Dynamic Time Warping, DTW) to compare the effect of segmentation, and the inertial sensor collects this action model data file as the matching standard, and the model data File in the file with the label, when the collection of new data, find the template data file has been marked in the segmentation point and the new data file in the corresponding point, you can segment the new data file, and Specific action segment data analysis motion performance.
This study evaluates the effect of DTW segment inertial sensor data in order to use DTW to compare the feasibility of segmentation of inertial sensor data. Therefore, this experiment is designed for inertial sensors to wear right wrist with three different modes The motion of the wrist is collected, and the wrist motion data are collected and compared with the intermediate point of the action switching time point and the action time, and the result of the segmentation error is evaluated by the method of absolute mean error, Cohen's d, T assignment.
Experimental results show that health and rehabilitation exercises than the error can be accepted, but the monster watch because of irregular movements, resulting in an absolute average error is too large. Therefore, in the data collection and analysis of somatosensory detectors, it is found that the differences in action alignment can be easily amplified in irregular movements, resulting in errors in dynamic time correction, The segmentation results are more accurate when the action is at the point of time.
|
|---|
