Dynamic Knee Valgus Detection During Squatting for Unsupervised Home Training
Verena Venek, Wolfgang Kremser, Harald Rieser, Michael Schechinger, Sonja Jungreitmayr (2019): Dynamic Knee Valgus Detection During Squatting for Unsupervised Home Training In: ECSS Book of Abstracts 2019
Abnormal movement patterns such as the dynamic knee valgus (DKV) can lead to serious joint problems. Pushing the knees too much inwards, e.g. during squatting, can occur particularly in unsupervised training. Thus, an affordable plug & play system should be developed to highlight DKV in real-time during exercise at home. Previous studies focused on assessing the squat quality using inertial measurement units (IMUs) to automatically categorize Single Leg Squat; however, using IMUs requires expert knowledge in setup and data gathering and is as such not yet ready for everyday use.
Therefore, we decided to investigate a camera-based system for DKV detection. After exploring the market, we decided on the Orbbec Persee, a 3D camera-computer system, and the Nuitrack skeleton tracking software. The proposed DKV detection algorithm uses the distance between the knee joints during the initial position and monitors it throughout the exercise. 41 subjects (23 female, 18 male, mean age 23.9±5.2 years) performed squats six times in front of the system. A sport scientist annotated the recordings to establish a ground truth about when DKV was occurring. Finally, the algorithm was evaluated by calculating the confusion matrix on both repetition- and subject-level.
Although the recall was over 70 % on both levels, the precision was about 40 % on subject-level. This means that although the algorithm detected most of the knee valgus positions correctly, it produces false positives as well.
The proposed DKV detection algorithm together with the Orbbec Persee were combined to a home training feedback system with telecoaching option. When an abnormal movement pattern is detected, a personal coach will be asked to verify the algorithm output. Since the system will be tested with over 100 people in the field, the DKV is only communicated to the personal coach if more than 20 % of the repetitions were classified as abnormal. Thus, less false positives can be obtained. Other exercises such as lunges were recorded of the 41 subjects and will be used for further analysis and optimization of the algorithm. In order to evaluate the effectiveness of the system on people’s knee stability, the Y-balance test and the functional movement screen (FMS) hurdle step will be examined before and after the six-month trial.