ECSS Paris 2023: OP-BM13
INTRODUCTION: The development of the rotational shot put technique analysis system aimed to enhance athletes performance through advanced artificial intelligence technologies. The system was designed to quickly capture and analyze biomechanical data of human motion, providing coaches and athletes with technical evaluations, thereby optimizing training effects. METHODS: System Architecture and Key Technologies The system primarily consists of two parts: the client and the server. Client includes a human-computer interaction module and a data acquisition and processing module, responsible for data input, display, and user interaction. Server includes a model invocation module and an offline training module, responsible for invoking deep learning models for pose estimation and data processing. Key technologies include a 3D human pose estimation algorithm, which utilizes an adaptive fusion method, combining epipolar expansion and triangulation techniques based on epipolar geometry constraints, improving adaptability to complex scenarios such as occlusion and enhancing pose estimation accuracy, a smoothing model, which improves the stability of pose sequences and reduces noise interference through smoothing processing, and a heartbeat mechanism, ensuring smooth communication between the client and server, guaranteeing real-time and accurate data processing. Specialized Action Technique Diagnosis The system is specifically designed for the rotational shot put athletes, providing detailed technical evaluation. Users can input training or competition data to generate diagnostic reports containing speed, amplitude, angle, time, and et al. Reports offer detailed technical analysis and feedback, helping coaches and athletes identify technical shortcomings. System Validation To verify the accuracy and reliability of the system, we set up the rotational shot put technique analysis system alongside the qualisys motion capture system in an experimental setting. Six athletes performed three rotational shot put throws, with both systems capturing the entire motion and analyzing kinematic parameters. Bland-Altman plot and ICC analysis were used to assess the consistency and correlation of measuring results between the two systems, respectively. RESULTS: Validation Results: The Bland-Altman plot showed high consistency in measuring results between the two systems. ICC analysis indicated a high correlation between the two systems. These results demonstrate that the rotational shot put technique analysis system has high accuracy and reliability, providing precise technical analysis outcomes for athletes and coaches. CONCLUSION: The rotational shot put technique analysis system is an accurate and reliable technical analysis tool, offering new methods and approaches for rotational shot put technique analysis. Through this system, coaches and athletes can conduct detailed technical evaluations and training optimizations, thereby improving athletic performance.
Read CV ZHAO DefengECSS Paris 2023: OP-BM13
INTRODUCTION: The analysis of shoulder kinematics during overhead movements can reveal important information for performance optimization and injury prevention in sports such as handball and tennis. Marker-based motion capture systems provide highly accurate joint angle measurements and are commonly used in sports science and clinical research. However, their use in field environments is limited due to long preparation and post-processing times. While markerless motion capture, like Theia 3D, presents a promising alternative, its accuracy for overhead shoulder kinematics remains unclear. Therefore, we aimed to compare shoulder kinematics during overhead movement tasks between Theia 3D and a marker-based motion capture system. METHODS: Twenty participants performed overhead movements including full arm scaption, abduction and flexion while synchronized markerless (Theia 3D, v2023.1) and marker-based (Qualisys) motion capture data were collected. Reflective markers were placed on landmarks of the upper extremities and trunk according to the International Society of Biomechanics (ISB) guidelines [1]. For both approaches data were processed with a low-pass Butterworth filter at 6 Hz cut-off and shoulder kinematics were reconstructed using the X-Z-Y sequence for scaption and abduction tasks and the Z-X-Y sequence for flexion tasks [2]. Coefficients of multiple correlation (CMCs) and root mean square errors (RMSEs) were calculated for the kinematic waveforms for each angle and movement task. RESULTS: In general, markerless and marker-based trajectories showed similar patterns for all movements, indicated by good to excellent CMCs (>0.75) for all angles and movement tasks. However, larger differences between systems were observed near the turning point of the movement, when the arms reached the maximum overhead position. Mean RMSEs were smallest for the abduction angle, ranging from 5.9° to 9.1°, followed by the flexion angle ranging from 8.7° to 12.1° and the internal rotation from 16.5° to 24.1°. CONCLUSION: While Theia3D v2023.1 showed general agreement in movement patterns compared to the marker-based system, notable differences in absolute joint angles were observed, particularly for the internal rotation angle. This suggests that Theia3D, in the tested version v2023.1, is not suitable for use in performance analysis and injury prevention in sports, where accurate measurements of overhead shoulder kinematics across three dimensions are essential. However, further refinement and training of the model on overhead movement tasks may enhance the accuracy of Theia3D for these tasks, making it a potent alternative for assessing shoulder kinematics in the future. References 1 Wu et al., J. Biomech, 2005 2 Lempereur et al., J. Biomech, 2014
Read CV Christoph KünzelECSS Paris 2023: OP-BM13
INTRODUCTION: Markerless motion capture (MMC) has emerged as a potential solution to motion capture in logistically challenging environments (e.g., outside a motion capture laboratory). As such, many studies compare MMC (e.g., Theia3D [1]) with conventional marker-based methods to understand agreement between systems. These studies often take place in a laboratory environment, where factors like lighting and camera position can be optimised for 3D coverage. However, many environments where MMC might be employed have suboptimal lighting or limited space for ideal camera placement (e.g., hospital or gymnasium). This study aimed to understand the effects of lighting and camera positioning on MMC performance, by synchronously comparing with a marker-based system. METHODS: Fifteen adults walked (3 km/h) and ran (12 km/h) on an instrumented treadmill (Gaitway3D; h/p/cosmos), whilst a 14-camera optoelectronic motion capture system (Oqus 7+, Qualisys) collected retroreflective marker trajectories [2]. Video data were collected synchronously with 12 high-speed cameras (100 Hz; Miqus, Qualisys). To compare the effect of lighting, data were collected under four randomised lighting conditions (1069, 692, 455, and 289 lux). Video cameras evenly surrounded the capture volume (SURR), but to compare the effect of camera position some cameras were removed post-collection to replicate a corridor (CORR) setup, and a setup with bias towards anterior views (FRONT) under one lighting condition (1069 lux). Videos were processed using Theia3D. Hip, knee, and ankle joint range of motion (ROM) was computed in all three planes with Visual3D and compared between lighting conditions and camera setups using marker data as a reference system. RESULTS: Significant interactions were found for all joints when lighting was changed, with less lighting leading to lower hip frontal ROM, higher knee and ankle frontal ROM, and higher ROM in all transverse angles (p < 0.05). Despite systematic differences between marker data and MMC, there were generally no interactions with lighting in the sagittal plane. There were significant differences between camera setups, particularly in frontal and transverse planes (p < 0.05). SURR and FRONT underestimated ROM compared with marker data, where CORR was in some cases more similar (p ≥ 0.05). CONCLUSION: Lighting and camera positioning need to be considered when using MMC, especially if kinematics in the frontal or transverse planes are of interest (e.g., in injury “screening” assessments). Equally, the translation of system comparison studies into the field-based adoption of MMC needs to be approached with caution, as external factors like lighting and camera position affect the difference between systems. Finally, if a motion capture environment has compromised lighting or limited space for camera setup, the accuracy of MMC might be limited. References: [1] Kanko, RM et al. (2021) J Biomech;127:110665 [2] Cappozzo, A et al. (1995) Clin Biomech;10:171-8
Read CV Josh WalkerECSS Paris 2023: OP-BM13