Design and optimization of intelligent teaching aid system for badminton: research on sports interaction experience

Real-time badminton stroke feedback requires low-latency, sequence-consistent recognition. A pipeline combining k-nearest-neighbour classification with Hidden Markov temporal sequencing was developed to produce time-stamped stroke labels suitable for live cueing and post-session review. Motion-capture data included 10,000 annotated strokes from 36 players (2015-2017, Chongqing, China). Features (angular velocity and jerk) were computed in 200 ms sliding windows with a 50 ms overlap. In five-fold cross-validation, KNN-HMM achieved 97.5% accuracy, 32 ms latency, and 90.8% streaming accuracy, with significant gains over standalone KNN. Findings support the deployment of stroke-level feedback under live constraints. Throughput confirmed real-time operation; generalisability beyond the recorded cohort remains untested externally.