Estimating Joint Kinetics During Baseball Pitching Using In-stadium Markerless Motion Capture

Markerless motion capture facilitates the collection of biomechanics data through machine-vision based approaches. In-stadium markerless motion capture has emerged as the primary modality for ball tracking and skeletal pose estimation of athletic movements at competitions. However, the criterion validity of baseball pitching kinetics using this markerless technology remains unclear. PURPOSE: To examine the equivalence of baseball pitching kinetics derived from in-stadium markerless, in-lab markerless, and marker-based motion capture modalities. METHODS: This analysis is based on a sample of 18 collegiate baseball players (height = 1.83 ± 0.02 m, mass = 86 ± 18 kg, fastball velocity = 84.3 ± 4.1 mph), all of whom provided written informed consent to participate in this study. Each participant pitched off a regulation mound while their segment motions were concurrently recorded with a 5-camera in-stadium markerless motion capture system (Hawk-Eye, Basingstoke, UK), 10-camera in-lab markerless motion capture system (Theia Markerless, Kingston, ON), and an 8-camera marker-based motion capture system (Qualisys, Göteborg, Sweden) at a sampling rate of 300 Hz. The concordance correlation coefficient (CCC) and a nested Bland-Altman analysis were used to assess the agreement and reliability of peak shoulder forces, peak shoulder rotation torques, and peak elbow valgus torques among modalities. RESULTS: Using the marker-based method as a reference, Theia and Hawk-Eye markerless modalities underestimated shoulder forces by 42 N and 37 N, respectively, with moderate reliability (CCC = 0.659). Theia underestimated shoulder and elbow torques by 16 Nm and 4 Nm (CCC = 0.426), respectively, while Hawk-Eye overestimated them by 33 Nm (CCC = 0.010). CONCLUSION: The in-stadium markerless system estimated pitching-relevant joint torques with larger biases, owing to error propagation from key point noise and deviations in segment definitions at the kinematics level.