Developments in Deep Learning for Low Light Object Detection: A Comprehensive Review of YOLO and YOLOv8

摘要

Object Detection (OD) is one among the challenging tasks in the Computer Vision (CV) field. Significant challenges, especially in Low-Light (LL) environments, are owing to the minimized visibility, increased noise, and poor contrast that often accompanies such conditions. Deep learning (DL)-centric techniques have taken over this field because of its rapid development. DL applications for LL object identification are numerous and include robotic activities at night, drone attacks, and reconnaissance and surveillance. Techniques still faced problems when used directly in LL, including missing detections and false positives although conventional OD techniques showed good results on datasets with typical illumination. Moreover, by the presence of small, dense, and obstructed objects, the models’ Detection Accuracy (DA) in LL is further reduced. Nevertheless, You Only Look Once (YOLO), which is also a DL algorithm, plays a crucial role in accurate Low-light Object Detection (LOD), and the updated version YOLOv8 enhances this capability even further. YOLOv8’s advanced neural network architecture allows for better Feature Extraction (FE), improved image quality, and higher DA in challenging Low-Light conditions (LLC). Thus, knowing the objective of the review paper helps to explore the DL growth and significance of YOLO and YOLOv8 for LOD