Edge Detection-Based Feature Extraction for the Systems of Activity Recognition.

Area of Science:

• Computer Science
• Artificial Intelligence
• Signal Processing

Background:

• Human Activity Recognition (HAR) systems face challenges with accuracy, particularly in noisy environments.
• Existing edge detection operators often fail to handle hostile conditions in activity frames.
• Effective feature extraction is crucial for robust HAR system performance.

Purpose of the Study:

• To design an adaptive feature extraction method based on edge detection for improved HAR.
• To address the limitations of existing operators in handling noisy activity frames.
• To enhance the signal-to-noise ratio (SNR) and localization of extracted features.

Main Methods:

• Developed an adaptive feature extraction method utilizing edge detection principles.
• Calculated edge direction considering non-maximum suppression for enhanced feature extraction.
• Processed frames to extract relevant information into feature vectors for classification.

Main Results:

• The proposed method demonstrates potential for better performance compared to other operators, especially with 'footstep function' like edges.
• The technique enlarges the product of signal-to-noise ratio (SNR) and localization.
• Evaluated performance on a depth dataset with thirteen distinct actions.

Conclusions:

• The proposed edge-based feature extraction method offers a promising approach for robust HAR.
• The adaptive nature of the method allows for better handling of noisy data.
• The feature vectors generated are suitable for feeding into classifiers for accurate activity recognition.