Segmentation Of Wushu Kicking Motion From Motion Capture Data Using MATLAB

Abstract

Motion segmentation plays a crucial role in analyzing complex martial arts movements, such as Wushu kicking, which often involve non-repetitive, high-variation actions. Traditional motion segmentation methods struggle to effectively capture transitions in such irregular movements. This study adapts an existing motion segmentation framework by Sonoda et al. (2008), using three key features2angular velocity, direction change, and volume change2to segment Wushu MoCap data. The research implements these methods on 3D motion capture data of Wushu kicks, with preprocessing steps carried out in MATLAB for visualization and analysis. The performance of each segmentation method is evaluated by comparing the results against ground truth data collected from experienced martial artists. Results show that angular velocity performs best, achieving a similarity score of 0.9574 and correctly matching 11 of 12 ground truth transitions. Direction change and volume change both score 0.8660, but each matches a different subset of transitions, highlighting the unique strengths and limitations of each approach. The integrated method, while offering broader coverage, did not improve over angular velocity alone, suggesting that simple fusion may propagate errors from individual methods. This work confirms that rotational motion is a reliable cue for detecting phase WUDQVLWLRQVLQPDUWLDODUWV7KHVWXG\¶VOLPLWDWLRQVLQFOXGHLWVUHOLDQFHRQDVLQJOH:XVKXNLFNLQJ sequence, the challenges of capturing subtle transitions using certain methods, and the fixed 100-frame tolerance window. Future work should focus on dynamically adjusting the tolerance window, refining threshold computation to mitigate jitter effects, expanding the dataset to include a wider variety of martial arts techniques, incorporating upper body motions, and exploring real-time applications in coaching systems. Additionally, comparisons with machine learning-based segmentation methods could further enhance the robustness and accuracy of the framework.