Integrating image enhancement with YOLO for improved low-light face detection

The paper introduced an innovative to enhancing face recognition performance in low-light environments using the You Only Look Once (YOLO) framework. Low light conditions pose a significant challenge to accurate facial identification, often degrading the reliability of conventional recognition systems. To address this issue, the study integrate image enhancement techniques, specifically Retinex and Zero-DCE, to strengthen the YOLO algorithm’s effectiveness. Through extensive experimentation, the YOLOv8m.pt model, optimised using Stochastic Gradient Descent (SGD) at the 30th epoch, emerged as the most effective configuration. The proposed system was rigorously evaluated for both speed and accuracy, demonstrating its suitability for real-time applications. Notably, the Zero-DCE method consistently outperformed Retinex, offering faster processing times and improved recall across most test images. Under optimal conditions, the YOLO-based system achieved a mean Average Precision (mAP) of 75.8%, a Precision of 76.5%, and a Recall of 66.4%, marking a substantial advancement in face recognition capabilities for security and surveillance in low-light scenarios. Future research might be aimed at expanding scope of the data to cover more diverse situations and creating high quality API to enable broader usage and continuous learning.