See Degraded Objects: A Physics-Guided Approach for Object Detection in Adverse Environments.

Area of Science:

• Computer Vision
• Image Processing
• Artificial Intelligence

Background:

• Object detection struggles in adverse environments due to degraded visibility and weakened features.
• Existing image enhancement methods introduce noise, hindering detector performance.

Purpose of the Study:

• To develop a physics-guided approach for robust object detection in adverse environments.
• To enable the detection of poorly visible objects by integrating physical knowledge.

Main Methods:

• Proposing a physics-guided object detection framework.
• Utilizing environment prior (e.g., atmospheric model) and frequency prior (amplitude spectrum).
• Implementing a physics-guided loss function with a novel weight item based on physical priors.

Main Results:

• The method effectively injects physical knowledge into detectors via backpropagation.
• Demonstrated state-of-the-art performance in synthetic and real foggy, and underwater scenarios.
• Successfully detects degraded objects that are otherwise nearly invisible.

Conclusions:

• The proposed physics-guided approach offers a straightforward and effective solution for object detection in adverse conditions.
• Physical priors and a guided loss function significantly enhance detection capabilities for degraded objects.
• The method achieves superior performance across diverse challenging environments.