Validity of Neural Networks to Determine Body Position on the Bicycle.

With the increased access to neural networks trained to estimate body segments from images and videos, this study assessed the validity of some of these networks in enabling the assessment of body position on the bicycle. Fourteen cyclists pedaled stationarily in one session on their own bicycles while video was recorded from their sagittal plane. Reflective markers attached to key bony landmarks were used to manually digitize joint angles at two positions of the crank (3 o'clock and 6 o'clock) extracted from the videos (Reference method). These angles were compared to measurements taken from videos generated by two deep learning-based approaches designed to automatically estimate human joints (Microsoft Research Asia-MSRA and OpenPose). Mean bias for OpenPose ranged between 0.03° and 1.81°, while the MSRA method presented errors between 2.29° and 12.15°. Correlation coefficients were stronger for OpenPose than for the MSRA method in relation to the Reference method for the torso ( = 0.94 vs. 0.92), hip ( = 0.69 vs. 0.60), knee ( = 0.80 vs. 0.71), and ankle ( = 0.23 vs. 0.20). OpenPose presented better accuracy than the MSRA method in determining body position on the bicycle, but both methods seem comparable in assessing implications from changes in bicycle configuration.