Autonomous modeling of repetitive movement for rehabilitation exercise monitoring

Table 1 Evaluation result from the proposed method. Area under the curve (AUC) and accuracy at optimal points

Exercise	Feature source	AUC	Optimal point accuracy*
Exercise	Feature source	AUC	Precision	Recall	F-1 score
RECT	Kinect alone	0.9299	0.8302	0.6111	0.7040
	Kinect + IMU	0.9753	0.9000	0.8750	0.8873
	Marker-based	0.9851	1.0000	0.8611	0.9254
COMB	Kinect alone	0.9624	0.9672	0.8082	0.8806
	Kinect + IMU	0.9924	0.9702	0.8904	0.9286
	Marker-based	0.9928	0.9306	0.9178	0.9241
TAP	Kinect alone	0.9733	0.8906	0.7917	0.8382
	Kinect + IMU	0.9985	1.0000	0.9583	0.9787
	Marker-based	0.9994	0.9351	1.0000	0.9664
POUR	Kinect alone	0.9282	0.8966	0.7222	0.8000
	Kinect + IMU	0.9827	0.9688	0.8611	0.9118
	Marker-based	0.9882	0.8590	0.9306	0.8933

*The optimal point is the point where the F-1 score is maximized (the orange points in Fig. 8)

ISSN: 1472-6947