Fig. 11
From: Autonomous modeling of repetitive movement for rehabilitation exercise monitoring
Anomaly detection results from sequences of COMB exercises with three simulated anomaly events recorded by Kinect+IMU system. These anomalous areas (highlighted in yellow) are manually labeled. Top: The extracted phase (without smoothing), and the approximately closest monotonically progressing (ACMP) phase sequence from the CDTW algorithm. The difference between them (red) is the anomaly scores from the phase fluctuation. Middle: The normalized reconstruction errors from different kinematic feature channels in different colors. Bottom: The normalized features used for the phase extraction