Related Experiment Video
Updated: Mar 3, 2026
04:23
A Swin Transformer-Based Model for Thyroid Nodule Detection in Ultrasound Images
Published on: April 21, 2023
2.4K
PoseShot: hybrid CNN-BiLSTM transformer model for free throw action recognition via pose analysis
Wei-Chun Hsu1, Cheng-Chi Lee2,3, Yong-Hsiang Lee4
1Graduate Institute of Biomedical Engineering, National Taiwan University of Science and Technology, Taipei, Taiwan.
Scientific Reports
|March 1, 2026
Summary
This study introduces PoseShot, a novel AI model for analyzing basketball free throw mechanics. It offers data-driven insights to enhance athlete performance by objectively assessing technique, moving beyond subjective evaluations.
Area of Science:
- Sports Science
- Artificial Intelligence
- Biomechanics
Background:
- Traditional basketball free throw analysis relies on subjective assessments, leading to bias and inconsistency.
- Existing human activity recognition models lack the granularity for fine-grained analysis of complex sports motions.
Purpose of the Study:
- To introduce PoseShot, a dual-channel hybrid CNN-BiLSTM-Transformer model for objective, data-driven analysis of basketball free throw mechanics.
- To enable fine-grained, phase-dependent analysis of individual free throw motions.
- To provide quantifiable insights for improving athlete performance.
Main Methods:
- Developed a dual-channel deep learning architecture integrating Convolutional Neural Networks (CNN) for spatial features, Bidirectional Long Short-Term Memory (BiLSTM) for temporal dynamics, and Transformer encoder for contextual understanding.
- Combined training footage with precise body posture angle calculations.
- Utilized a hybrid model incorporating CNN, BiLSTM, and Transformer for motion analysis.
Main Results:
- PoseShot achieved high performance metrics: F1-score of 95.76%, precision of 95.72%, and recall of 95.80%.
- Outperformed established architectures like DenseNet, Swin Transformer, and Vision Transformer in analyzing complex throwing motions.
- Identified key biomechanical determinants of successful free throws, offering actionable guidance for athletes and coaches.
Conclusions:
- PoseShot provides accurate, objective motion analysis, bridging the gap between subjective evaluation and advanced analytics.
- The model offers a transformative tool for sports performance analysis, revolutionizing basketball training.
- Data-driven insights from PoseShot facilitate posture optimization and action consistency for enhanced athletic performance.