Bio-Inspired Dark Adaptive Nighttime Object Detection.

Area of Science:

• Computer Vision
• Biomimicry
• Automotive Safety

Background:

• Nighttime object detection is crucial for automotive safety, but current systems struggle with poor lighting.
• Existing solutions often rely on expensive sensors like LiDAR and thermal imaging.
• Pedestrian anti-collision systems show reduced effectiveness at night, as per 2022 IIHS research.

Purpose of the Study:

• To develop a cost-effective 2D image-based approach for enhanced nighttime object detection.
• To simulate biological dark adaptation mechanisms for improved low-light performance.
• To create a robust testing environment for evaluating object detection across various lighting conditions.

Main Methods:

• Inspired by biological dark adaptation, simulating pupil and photoreceptor functions.
• Utilized image fusion and gamma correction for training deep neural networks.
• Developed a simulated environment with a dynamic range from 0 lux to high brightness.

Main Results:

• The developed dark adaptation model improved mean average precision (mAP) by 1.5-6% over traditional models.
• Achieved effective object detection in both twilight and nighttime conditions.
• Demonstrated a novel high dynamic range testing methodology.

Conclusions:

• The proposed low-cost 2D imaging method offers a significant improvement for nighttime object detection.
• The model's ability to function in varied low-light conditions presents academic novelty.
• Future integration with smart car lighting can further enhance detection accuracy and safety.