PoseShot: hybrid CNN-BiLSTM transformer model for free throw action recognition via pose analysis.

The evaluation of basketball free throw techniques has traditionally relied on subjective assessments, which introduces inherent biases and inconsistencies in performance analysis. This study presents PoseShot, a novel dual channel hybrid CNN-BiLSTM-Transformer Model that facilitates the comprehensive analysis of free throw mechanics with data-driven insights. Unlike conventional human activity recognition that focuses on coarse activity labels, PoseShot is designed to analyze fine-grained, phase-dependent mechanics within a single basketball free throw motion. The proposed framework integrates training footage with precise body posture angle calculations via a dual-channel deep learning architecture to enable the capture of subtle technical variations. Our innovative approach synthesizes convolutional neural networks (CNN) for spatial feature extraction, bidirectional long short-term memory (BiLSTM) for temporal sequence processing, and a transformer encoder for enhanced contextual understanding of motion dynamics. The model demonstrates exceptional performance, achieving an F