Document Type



Velocity based training (VBT) is a promising method to quantify and direct resistance training. Recent advances in computer science have opened the way for low-cost methods to measure VBT using video data from a smartphone. This work introduces and analyses the feasibility of a computer vision-based Python application in tracking barbell kinematics during VBT, compared against Vicon data during the back squat in one subject. As input into the algorithm, sagittal-plane video data is needed with the barbell plate in focus. Time of the concentric part of the squat and vertical barbell displacement are then automatically tracked using OpenCV libraries. The time parameter was accurately assessed using two different OpenCv Tracker, the KCF (r=0.83, SEE=0.02s) and the MOSSE (r=0.81, SEE=0.02s) tracker, respectively. For the vertical displacement, a lower accuracy was obtained using KCF (r=0.36, SEE=0.02m) and MOSSE (r=0.62, SEE=0.01m). Tracking errors could be explained by the camera set-up, as well as differences in frame rates between the video and the Vicon data. It might be possible to correct these errors in future work using machine learning techniques. This pilot study shows the feasibility of a computer vision-based Python application to measure barbell kinematics in a low-cost manner and might play a part towards advancing VBT monitoring technologies for widespread use.