Abstract
This research presents a Knowledge-Based Coaching System (KBCS) designed to modernize professional sports training through the integration of Artificial Intelligence (AI) and Human-Computer Interaction (HCI). By digitizing the "tacit knowledge" of expert coaches, the system provides a scientific framework for evaluating and instructing college athletes. Applied specifically to table tennis, the KBCS utilizes a hybrid motion capture approach—combining optical and inertial data—to achieve high-precision pose restoration and trajectory analysis. Experimental results demonstrate that the system effectively maintains trajectory continuity and provides accurate skill level assessments, significantly enhancing student engagement and technical performance compared to traditional coaching methods.
Introduction
The rapid development of digital technology has transformed many fields, yet professional sports coaching often remains tethered to subjective human experience. This "tacit knowledge," while valuable, is difficult to scale and lacks the quantitative precision required for optimal athletic development. In college sports education, there is a growing need for automated systems that can provide personalized, data-driven guidance. This paper addresses this gap by proposing the Knowledge-Based Coaching System (KBCS), a framework that applies educational psychology principles to autonomous sports training.
System Architecture: The KBCS
The KBCS is built upon four core modules: domain knowledge, an athlete model, sports evaluation, and a controller. The system functions by comparing an athlete's real-time performance against a "Reference Library" of professional standards.
As shown in Figure 2, the system integrates practice and reference libraries to evaluate athletic performance. When a discrepancy is detected between the student's motion and the ideal form, the system selects appropriate feedback from an "Instruction Library" to guide the athlete toward improvement.
Technical Implementation: Hybrid Motion Capture and Pose Correction
To ensure high-fidelity data, the researchers employed a dual-track motion capture strategy. An optical system (Nokov) tracks the high-speed trajectories of the ball and racket, while an inertial system (Legacy) monitors the athlete's skeletal poses. A significant challenge in inertial motion capture is "jitter" and position drift. To mitigate this, the authors developed a pose correction algorithm that uses the racket's position—captured accurately via optics—as a constraint to correct the human skeleton model in real-time. Additionally, a temporal-spatial feature algorithm was implemented for data mosaicking, ensuring continuous and smooth trajectories even when segments of motion data are unmarked.
The HCI Training Loop
The heart of the KBCS is its closed-loop Human-Computer Interaction (HCI) training process. This process mimics the adaptive teaching style of a human coach by adjusting difficulty and feedback based on the athlete's immediate performance.
The controller, as depicted in Figure 3, manages motion capture, feature extraction, and sports evaluation. If an athlete fails to meet specific performance thresholds, the system triggers a "Skill Update" command, providing targeted instructional feedback to correct the error.
Experimental Results and Evaluation
The system was tested using three table tennis athletes of varying skill levels. The mosaicking algorithm successfully maintained speed and position continuity across 500-frame flight paths. Analysis of serve trajectories revealed that slight-spin and back-spin serves achieved higher accuracy than top-spin serves, and that lower serve speeds generally correlated with higher landing precision.
Furthermore, a questionnaire survey of 30 college athletes and 6 coaches indicated high levels of subjective acceptance. The KBCS scored 3.72 in "Acceptance" and 3.54 in "Training Enthusiasm," outperforming traditional training methods. The system's skill level classification was also found to be consistent with actual participant performance, proving its reliability in small-sample environments.
Conclusion
The KBCS represents a significant step toward the digitalization of sports education. By fusing AI with HCI and hybrid motion capture, the system provides a robust, objective alternative to traditional coaching. Future work will focus on reducing system latency, simplifying the user interface for broader adoption in physical education classes, and expanding the framework to other sports involving moving target detection.
Summary updated