Physiotherapy Exercise Error Detection

The usage of wearables is increasing in clinics for monitoring patients. One such application of wearables is to monitor and evaluate physiotherapy exercises. The usage of sensors alone does not always suffice due to drift in the readings of the sensors. Hence, we need to fuse readings from sensors in order to obtain a more accurate reading for joint angles.

We use a simple optimization method to fuse the readings obtained from two cameras and 10 IMUs. A SMPL model is obtained, that reflects the joint angles of the person performing the exercise.

Joint angles extracted from the model are used to classify the exercise into error and no error.

1 / 7
Fitting pipeline
The fitting pipeline
2 / 7
Sensor fusion results
Fusing more data improves our estimate of joint angles
3 / 7
Comparison of modalities
Modalities used in the comparison
4 / 7
Classifier results
Results obtained from the classifier
5 / 7
Transform comparisons
Comparison of different transformations used
6 / 7
Sliding window visualization
A visualization of the sliding window method
7 / 7
Sliding window gif
A visualization of the sliding window method