Seeing Pedestrian in the Dark via Multi-Task Feature Fusing-Sharing Learning for Imaging Sensors

Area of Science:

• Computer Vision
• Machine Learning

Background:

• Pedestrian detection is vital for autonomous systems but struggles in low-light conditions.
• Existing methods often fail to perform effectively in uncontrolled, low-illumination environments.

Purpose of the Study:

• To develop a novel pedestrian detection algorithm specifically for low-light environments.
• To address the limitations of current methods in adverse lighting conditions using multi-task learning.

Main Methods:

• A unique multi-task learning approach with feature-level fusion and a sharing mechanism.
• The method integrates an image relighting subnetwork, a pedestrian detection subnetwork, and a fusion module.
• This architecture enhances low-light image quality and simultaneously learns robust detection features.

Main Results:

• The proposed approach significantly improves pedestrian detection performance on low-light images.
• Experimental results demonstrate consistent and substantial performance gains.
• The feature-level fusion effectively boosts both image relighting and detection.

Conclusions:

• The novel multi-task learning algorithm effectively tackles pedestrian detection challenges in low-light scenarios.
• This method offers a significant advancement for computer vision systems operating in real-world, variable lighting conditions.