Fast highway abandoned object detection via block-based multi-group foreground extraction.

Area of Science:

• Computer Vision
• Artificial Intelligence
• Traffic Safety Engineering

Background:

• Abandoned objects on highways present a significant traffic accident risk.
• Existing detection technologies face hardware and real-time processing limitations in complex highway environments.

Purpose of the Study:

• To propose a novel Universal foreground extraction framework (UBMG) for robust abandoned object detection on highways.
• To address false alarms caused by illumination changes, traffic, or road markings.

Main Methods:

• Developed a framework combining Block-based frame selection and Multi-Group foreground Detection (UBMG).
• Implemented block preprocessing, adaptive size filtering, and static target matching for candidate extraction.
• Utilized a candidate verification strategy including traffic target elimination, road noise elimination, and trajectory discrimination.

Main Results:

• The UBMG framework demonstrated robust and real-time detection of abandoned highway objects.
• Achieved superior performance compared to state-of-the-art methods on the newly established HAO dataset.
• Showed good performance on the public ABODA dataset.

Conclusions:

• The UBMG framework effectively alleviates false alarms and enhances detection accuracy for highway abandoned objects.
• The framework integrates seamlessly with existing algorithms and shows high efficiency in practical applications.
• The developed HAO dataset provides a valuable resource for evaluating highway abandoned object detection systems.