LLOD:A Object Detection Method Under Low-Light Condition by Feature Enhancement and Fusion

Abstract

In this paper, we propose a novel approach to achieve our goal of implementing object detection under low-light conditions by incorporating several innovative components. First, we design a new feature fusion unit that enables semantic features to better align with target inspection characteristics. Second, we introduce a novel low-light enhancement encoder unit to augment the semantic features of low-light images. Third, due to the limited availability of large-scale datasets for low-light scenes, we train an enhancement model first, which can effectively assist object detection in low-light conditions through feature enhancement. Our method demonstrates promising results in addressing the challenges of object detection under poor lighting conditions, providing a valuable contribution to the field of computer vision and enhancing the performance of object detection tasks in low-light environments.