Detection and evaluation of physical therapy exercises from wearable motion sensor signals by dynamic time warping

We develop an autonomous system to detect and evaluate physical therapy exercises using wearable motion sensor units. We propose an algorithm based on the dynamic time warping (DTW) dissimilarity measure to detect the occurrences of one or more exercise types in the recording of a physical therapy session. The algorithm evaluates the exercises as correctly/incorrectly performed, identifying the error type, if any. To evaluate the algorithm's performance, we record a data set consisting of one template execution and 10 test executions of each of the three execution types of eight exercises performed by five subjects. We thus obtain a total of 120 and 1,200 exercise executions in the training and test sets, respectively. The test signals also contain idle time intervals. The proposed algorithm detects 1,125 executions in the whole test set, where 8.58% of the 1,200 executions are missed and 4.91% of the idle time intervals are incorrectly detected as executions. The accuracy is 93.46 % for exercise classification only and 88.65 % for simultaneous exercise and execution type classification. To test the behavior of the system in case of unknown movements, the algorithm is executed for each exercise by leaving out the templates of that exercise, resulting in only 10 false alarms out of 1,200 executions. This demonstrates the robustness of the system against unknown movements. The proposed system may be used both for estimating the intensity of a physical therapy session and for evaluating executions of an exercise to provide feedback to the patient and the physical therapy specialist.