A Mobile Feedback Framework for Pose-Based Cricket Fielding Technique Analysis

Abstract

Cricket fielding analysis is a fine-grained sports action recognition problem because several techniques contain similar upper-body configurations but differ in posture transition, hand position, and movement timing. This paper proposes a mobile feedback framework for pose-based cricket fielding technique analysis. The framework connects video capture, prototype processing, pose extraction, fixed-length landmark representation, hybrid CNN–LSTM inference, confidence reporting, and recommendation delivery. MediaPipe Pose represents each video as a temporal landmark sequence, and the classifier recognizes five techniques: High Catch, Orthodox Cup, Reverse Cup, Short Barrier, and Long Barrier. The prototype is evaluated through functional, timing-related, and model-performance evidence. Beyond reporting model accuracy, the paper clarifies the architectural bridge between isolated action-recognition models and deployable mobile feedback systems, while identifying the additional ablation, deployment-latency, biomechanical-feedback, and user-study evidence required for stronger empirical validation.