Evaluation of a markerless motion capture system in a Premier League Football Academy
Coyne, Lara
Coyne, Lara
Loading...
Publication Date
2021-08-17
Type
Thesis
Downloads
Citation
Abstract
Current practice of movement analysis in elite youth football Academies is largely qualitative and confounded by bias. This thesis aimed to investigate a novel Markerless motion capture system that attempts to infer movement without markers. A dearth of research exists on markerless systems. Marker-based motion capture systems are the established method of measuring movement. Limitations of marker-based system due to soft tissue artefact, lengthy processing, and set up time exist. A lack of consensus is evident in the literature around how marker-based systems attempt to reduce this error, particularly concerning marker set selection and the biomechanical model used (3DOF or 6DOF) for large amplitude tasks. Study one considered four marker sets using both models to determine knee kinematics in all three planes (n=10) during the overhead squat, using RMSE, CCF and repeated measures ANOVA analysis. The RMSE were between 1.02 and 4.19 degrees and the CCF were excellent (>0.99) in the sagittal plane, while reduced in the non-sagittal planes. The findings revealed significant differences between the marker sets in the frontal and transverse planes (p<0.05), with an additional significant interaction for the marker set and DOF found in the transverse plane. The Anterior Partial Cluster using both 3DOF and 6DOF emerged as optimal. Study one also aimed to ensure the integrity of the marker-based system data in preparation for study two. Study two concurrently examined the agreement of the markerless and the marker-based systems between sessions, considering the lower limb peak angles and range of motion during the overhead squat and lunge (n=9) using the Bland-Altman method. Overall, the markerless system tended to overestimate the derived kinematics as revealed by the consistent negative mean bias over both tasks and sessions. The agreement intervals varied across all outcomes and planes, marginally wider for the 6DOF marker-based system and 3DOF markerless system comparison. Larger mean bias and poorer agreement was noted for the hip segment that may require an offset. The XXIV consistent mean bias across outcomes, limbs, between sessions, suggests a degree of repeatability between the systems according to each plane, however the magnitude of the rotations suggest caution. The current iteration of the markerless system, without calibration using the mean bias, cannot be considered exchangeable with the marker-based system. There is evidence of proportional bias during the lunge task, in particular in the transverse plane, that would require further examination. Study three considered the practical application of the markerless system for the determination of the lower limb mean peak kinematic differences at three season time points (n=27) during the overhead squat, lunge and vertical jump landing. A linear mixed effects model compared the repeated measures over time during three tasks. A post hoc analysis revealed significant differences during the three tasks over many of the right leg outcomes, predominantly occurring between pre-season start (T1) to mid-season (T3) and pre-season end (T2) to mid-season (T3) (p<0.05). The performance outcomes such as the overhead squat depth, vertical jump landing absorption depth, lunge distance demonstrated significant differences (p<0.05) between (T1 vs T3) and (T2 vs T3) testing sessions respectively. The vertical jump peak landing kinematics revealed significant increases for the hip (approximately 13 degrees) and the knee (between 5 and 7 degrees) in the sagittal plane. Many questions remain unanswered regarding the markerless system measurement error that requires addressing in future work. As markerless system algorithms evolve, the advancement in the quantification of kinematics for practical applications will occur, where ongoing research is warranted.
Funder
Publisher
NUI Galway