Enhanced human hitting movement recognition using motion history image and approximated ellipse techniques

Recognition of human hitting movement in a more specific context of sports like boxing is still a hard task because the existing systems use manual observation which could be easily flawed and highly inaccurate. However, in this study, an attempt is made to present an automated system designed for this purpose to detect a specific hitting movement commonly known as a punch using video input and image processing techniques. The system employs Motion History Image (MHI) to model trajectories of motions and combine them with other parameters to reconstruct movements which tend to have a temporal component. Thus, CCTV cameras set at different positions (front, back, left and right) enable the system to identify several types of punches including Jab, Hook, Uppercut and Combination punches. The most important aspect of this work is the proposal of MHI and the Ellipse approximation which is quicker in the integration of both than other sophisticated systems which take a considerable duration in computations. Therefore, the system classifies C_motion, Sigma Theta, and Sigma Rho parameters to distress hitting from non-hitting movements. Evaluation on a dataset captured from multiple viewpoints establishes that the system performs well achieving the goal of 93 percent when detecting both the hitting and the non-hitting motion. These results demonstrate the system’s superiority to the system based such detection methods. This study paves the way for other applications in realtime such as sports analysis, security surveillance, and healthcare requiring greater efficiency in and accuracy of human movement assessment. The focus of future work may be in the direction of improving the recognition of slower movements, also modifying the system for more dynamic conditions in the future.