Related Experiment Video
Updated: Mar 4, 2026
06:37
Author Spotlight: Addressing Technical and Subjective Challenges in Measuring Classroom Attention
Published on: December 15, 2023
5.5K
Application of LSTM-CNN in skiing action recognition under artificial intelligence technology
Wenhao Zhang1, Liang Xu2, Lei Wang3
1Ice and Snow Industry College, Jilin University of Physical Education, Changchun, 130022, Jilin, China.
Scientific Reports
|March 2, 2026
Summary
This study introduces a deep learning model for accurate skiing action recognition, even in challenging conditions like poor lighting and background clutter. The model achieves high precision and recall, offering a robust solution for intelligent sports analysis.
Area of Science:
- Computer Vision
- Artificial Intelligence
- Sports Analytics
Background:
- Automatic action recognition in sports is challenging due to complex environmental factors.
- Skiing action recognition faces difficulties with background interference, lighting variations, and self-occlusions.
Purpose of the Study:
- To enhance the accuracy of automatic skiing action recognition in complex scenarios.
- To develop a deep learning model that effectively integrates visual perception for improved performance.
Main Methods:
- A two-stream three-dimensional convolutional network-bidirectional long short-term memory (C3D-BiLSTM) architecture was employed.
- The model utilizes Red Green Blue (RGB) and saliency perception streams with optical flow.
- A learnable weighted fusion method integrates features, followed by Bidirectional Long Short-Term Memory (BiLSTM) for sequence modeling.
Main Results:
- The proposed C3D-BiLSTM model achieved superior performance over baseline models with 92.8% precision, 91.9% recall, and 0.923 F1-score.
- Ablation studies confirmed the effectiveness of the BiLSTM, saliency stream, and weighted fusion.
- The model demonstrated over 85% accuracy in cross-scenario and cross-athlete tests, with good noise stability.
Conclusions:
- The model effectively recognizes complex skiing movements by leveraging appearance and motion information.
- The proposed approach offers new technical methods for intelligent sports analysis.
- The C3D-BiLSTM model provides a stable and accurate solution for skiing action recognition.