A parallel CNN architecture for sport activity recognition based on minimal movement data.

Area of Science:

• Computer Science
• Biomedical Engineering
• Artificial Intelligence

Background:

• Human Activity Recognition (HAR) using sensors is advancing rapidly.
• AI is increasingly applied to complex scientific research problems.
• Accurate sports activity identification benefits all age groups.

Purpose of the Study:

• To propose a novel approach for categorizing sports activities using motion sensor data.
• To leverage parallel Convolutional Neural Networks (CNN) and machine learning for enhanced HAR.
• To improve the precision of sports activity classification.

Main Methods:

• Data preprocessing and normalization of sensor data.
• Feature extraction using Discrete Wavelet Transform (DWT) and Short-Time Fourier Transform (STFT).
• Parallel CNN models for motion feature extraction, followed by Random Forest classification.

Main Results:

• The proposed method achieved a mean precision of 99.61% on the DSADS dataset.
• Parallel CNNs effectively constructed motion features from diverse signal representations.
• Random Forest classification accurately identified sports activities based on merged features.

Conclusions:

• The novel HAR methodology demonstrates high effectiveness in sports activity classification.
• The parallel CNN and machine learning approach offers a robust solution for motion sensor data analysis.
• This technique holds potential for future AI-driven scientific research and applications.