Heterogeneous Fusion of Camera and mmWave Radar Sensor of Optimizing Convolutional Neural Networks for Parking Meter...

Area of Science:

• Computer Vision
• Sensor Fusion
• Machine Learning

Background:

• Street parking detection is challenging due to environmental factors like traffic, shadows, and weather.
• Traditional methods struggle with reliability in adverse conditions such as rain, fog, dust, and glare.

Purpose of the Study:

• To propose a novel heterogeneous fusion of convolutional neural networks for robust smart parking meter systems.
• To enhance parking region detection accuracy under diverse and difficult environmental conditions.

Main Methods:

• A heterogeneous fusion approach combining RGB camera and active mmWave radar sensor data.
• Utilizing convolutional neural networks for individual sensor data processing and fusion.
• Implementation on a GPU-accelerated embedded platform (Jetson Nano) for real-time performance.

Main Results:

• The proposed method accurately detects parking regions even in adverse conditions like rain, fog, dust, snow, and glare.
• Experimental results demonstrate an average accuracy of 99.33% for the heterogeneous fusion method.
• Real-time performance was achieved through heterogeneous hardware acceleration.

Conclusions:

• The developed heterogeneous fusion CNN model offers a highly accurate and robust solution for smart parking meters.
• The system effectively overcomes limitations of single-sensor approaches in challenging outdoor environments.
• This technology has the potential to significantly improve automated parking management systems.