Improving Radar Human Activity Classification Using Synthetic Data with Image Transformation.

Area of Science:

• Radar signal processing
• Machine learning applications
• Human activity recognition

Background:

• Machine learning (ML) is state-of-the-art in radar signal processing for tasks like human activity classification.
• Training ML models is challenging due to limited and non-diverse radar datasets.
• Existing methods struggle with the scarcity of real-world data for robust classification.

Purpose of the Study:

• To introduce an algorithm that enhances human activity classification accuracy using radar.
• To reduce the dependency on limited source datasets for ML model training.
• To leverage synthetic data generation for improved radar-based activity recognition.

Main Methods:

• Developed Radar Activity Classification with Perceptual Image Transformation (RACPIT) algorithm.
• Utilized a human radar reflection model based on video-captured motion for synthetic data generation.
• Implemented an image transformation network to align synthetic data with real radar data.
• Trained a Convolutional Neural Network (CNN) using augmented datasets for activity classification.

Main Results:

• Achieved up to a 20% increase in classification accuracy.
• Demonstrated the effectiveness of synthetic data augmentation.
• Reduced the need for collecting additional real-world radar data.
• Showcased the RACPIT algorithm's capability to improve radar-based activity classification.

Conclusions:

• The RACPIT algorithm significantly improves human activity classification accuracy in radar systems.
• Synthetic data augmentation, guided by perceptual image transformation, is a viable strategy to overcome data limitations.
• This approach offers a cost-effective and efficient method for enhancing radar-based activity recognition without extensive data collection.