Automatic Detection Method for Black Smoke Vehicles Considering Motion Shadows.

Area of Science:

• Environmental Science and Engineering
• Computer Vision and Artificial Intelligence
• Atmospheric Pollution Monitoring

Background:

• Mobile source pollutants, particularly vehicle tailpipe emissions, are significant contributors to air pollution.
• Motion shadows from vehicles visually mimic black smoke, complicating accurate detection.
• Traditional methods struggle with differentiating shadows from emissions, hindering effective pollution control.

Purpose of the Study:

• To develop an accurate and efficient method for detecting black smoke-emitting vehicles.
• To address and overcome the challenge of motion shadow interference in black smoke detection.
• To enable real-time monitoring of vehicle emissions for environmental protection.

Main Methods:

• Utilized YOLOv5s model for initial detection and localization of vehicles, motion shadows, and black smoke.
• Employed Simple Linear Iterative Clustering (SLIC) for superpixel image segmentation of detected objects.
• Developed a lightweight MobileNetv3 network (Y-MobileNetv3) for segmentation-classification, integrating shadow awareness.

Main Results:

• The Y-MobileNetv3 model achieved a high accuracy rate of 95.17% in detecting black smoke vehicles.
• Demonstrated a significant 4.73% accuracy improvement over models that do not account for motion shadows.
• Achieved a rapid average single-image inference time of only 7.3 milliseconds, enabling real-time application.

Conclusions:

• The proposed segmentation-classification approach effectively overcomes motion shadow interference in black smoke detection.
• The Y-MobileNetv3 model offers a robust solution for accurate and real-time identification of polluting vehicles.
• Superpixel segmentation aids in detecting even trace amounts of black smoke, enhancing emission monitoring capabilities.