A new approach for physical human activity recognition based on co-occurrence matrices.

Area of Science:

• Computer Science
• Signal Processing
• Biomedical Engineering

Background:

• Human Activity Recognition (HAR) is crucial for monitoring and understanding human actions.
• Sensor-based HAR technologies are increasingly important in daily life applications.
• Existing methods require effective feature extraction from sensor signals.

Purpose of the Study:

• To propose a new feature extraction method for sensor-based HAR.
• To adapt image processing techniques for one-dimensional sensor signals.
• To evaluate the effectiveness of the proposed method in activity recognition tasks.

Main Methods:

• A novel approach inspired by the Gray Level Co-Occurrence Matrix (GLCM) was developed for 1D signals (1D-GLCM).
• Heralick features were extracted from the co-occurrence matrix generated by 1D-GLCM.
• The method was tested on datasets from accelerometer, gyroscope, and magnetometer sensors.

Main Results:

• The proposed method achieved high success rates of 96.66% and 93.88% on two different datasets.
• The extracted Heralick features proved effective for HAR.
• The approach demonstrated strong performance in various activity recognition scenarios.

Conclusions:

• The novel 1D-GLCM-based feature extraction method is highly effective for HAR.
• This approach offers a promising technique for classifying diverse sensor signals.
• The study highlights the potential of adapting image processing concepts to signal analysis for HAR.