Detection of Aerobics Action Based on Convolutional Neural Network

Area of Science:

• Computer Vision
• Machine Learning
• Sports Science

Background:

• Accurate aerobics action detection is crucial for performance analysis and training.
• Existing methods like Faster R-CNN face challenges in capturing multiscale features effectively.

Purpose of the Study:

• To enhance the accuracy of aerobics action detection.
• To address limitations in current detection models by improving multiscale feature extraction.

Main Methods:

• The study proposes an aerobics action detection method based on improved multiscale characteristics.
• It utilizes the Feature Pyramid Network (FPN) integrated with Faster R-CNN to extract image features.
• This approach converts low-level semantic information into high-resolution, deep-level semantic information.

Main Results:

• The proposed method reduced the neural network's loss function to 0.2.
• Achieved a detection accuracy of 96.5%, outperforming other existing methods.
• Demonstrated the feasibility and effectiveness of the enhanced feature extraction technique.

Conclusions:

• The integration of FPN with Faster R-CNN significantly improves aerobics action detection.
• The method effectively captures and utilizes multiscale features for higher accuracy.
• This research provides a viable solution for more precise aerobics action recognition.