A Nighttime Vehicle Detection Method with Attentive GAN for Accurate Classification and Regression

Area of Science:

• Computer Vision
• Artificial Intelligence
• Autonomous Driving Systems

Background:

• Vehicle detection is crucial for autonomous driving systems.
• Nighttime conditions present significant challenges due to poor visibility and lighting.
• Existing methods struggle with feature extraction in low-light environments.

Purpose of the Study:

• To propose a high-accuracy vehicle detection algorithm for nighttime scenes.
• To enhance vehicle features in nighttime images using an improved Generative Adversarial Network (GAN).
• To improve detection accuracy through advanced regression and classification techniques.

Main Methods:

• Utilized an Attentive Generative Adversarial Network (GAN) for nighttime image feature enhancement.
• Employed multiple local regression for predicting bounding box offsets.
• Integrated an improved Region of Interest (RoI) pooling method with Faster Region-based Convolutional Neural Network (R-CNN).
• Incorporated cross-entropy loss to refine classification accuracy.

Main Results:

• The proposed algorithm demonstrated effective contribution to nighttime vehicle detection accuracy.
• Experimental results showed superior performance compared to state-of-the-art detectors.
• The method successfully enhances subtle vehicle features in challenging nighttime conditions.

Conclusions:

• The developed algorithm significantly improves vehicle detection in nighttime driving scenarios.
• The Attentive GAN and refined Faster R-CNN architecture offer a robust solution for autonomous driving.
• This research advances the reliability of autonomous systems in adverse lighting conditions.