“Low-Light Image Enhancement with YOLO for Night Object Detection”

Abstract

Detection of objects in low-light or night time conditions is still a major drawback of computer vision systems as a result of inadequate lighting, contrast, noise and distortion of visual information. Such constraints usually decrease precision and dependability of real-time object detect models applied in processes of surveillance, self-driving, and surveillance. The proposed paper is a hybrid model whereby a combination of low-light image enhancement methods and You Only Look Once (YOLO) object detection model is combined in order to enhance detection in the night-time environment. The first stage involves image enhancement algorithm which enhances the image based on the light levels to improve the brightness, contrast and create structure of the image as well as reduces noise. The images of the enhanced are then used through the object detection model that is based on the YOLO (so that the objects are accurately localized and identified even in harsh light conditions). The suggested solution should increase the level of feature visibility and allow the detection model to identify the significant details in the already ambiguous images. Through experimental assessment, it is protested that image enhancement in combination with YOLO has a greater detection accuracy, precision and recall than traditional object detection frameworks implemented on low-light photos without image enhancement. The findings suggest that the presented framework can be applied to the actual night-time detection scenarios and serve to enable the advanced applications of vision based on the need to detect objects in low light reliably to ensure future application in the advanced systems.