Exploring the accuracy of OpenCap for three-dimensional analysis of cricket bowling

This study explored the accuracy of pose estimation software, OpenCap, against a marker-based motion capture system for cricket bowling. Ten participants (nine male, one female; seven pace, three spin; nine right-arm, one left-arm; age = 22.8 ± 4.1 years; height = 181.8 ± 7.0 cm; body mass = 82.54 ± 7.6 kg) bowled 48 deliveries with data simultaneously collected by OpenCap and marker-based motion capture. Shoulder (arm elevation plane, arm elevation angle and axial rotation), elbow (flexion), trunk (rotation, extension and lateral flexion), knee (flexion) and ankle (dorsiflexion) joint kinematics were extracted from back foot contact to ball release for each ball bowled. Average (± standard deviation) root mean squared error (RMSE) for each joint angle between OpenCap and motion capture was calculated. 95% limits of agreement were calculated for joint kinematics at back foot contact, front foot contact and ball release events. Of the 473 total trials completed, 217 trials were deemed successful across both OpenCap and motion capture. OpenCap had an average RMSE of 17.61° (± 7.72°) across all joint angles. OpenCap was able to most accurately determine knee kinematics (RMSE = 7.87 ± 2.10°) whilst upper limb kinematics were the least accurate (elbow RMSE = 22.71 ± 7.31°, arm elevation RMSE = 17.59 ± 3.78°, arm elevation plane RMSE = 28.92 ± 5.32°, shoulder axial rotation RMSE = 28.54 ± 7.86°). The relatively large error in upper limb kinematics and number of unsuccessful trials captured makes it challenging at present to recommend OpenCap for use in field-based analysis of cricket bowling kinematics.