Enhanced Real-Time Face Detection in Challenging Environments: Leveraging YOLOv8 for Partially Occluded and Low-Light Conditions

Face recognition technology is considered a crucial component of modern surveillance frameworks; however, face recognition under unfavourable conditions, including partial obstruction and low-light scenarios, is still a prominent challenge. This paper analyzes how the deep learning model YOLOv8 can be used to solve these issues. The model was tested with a customized set of photos illustrating faces with different levels of occlusion (such as masks, helmets and scarves) and lighting. The researcher has classified the data into three categories 80% of the data is used in training, 10% of the data is used in testing, and 10% of the data is used in validation. The training process was taken 150 times, with critical measures such as 93% precision, 88.7% recall, an F1-score of 0.90, and a mean average precision (mAP@50) of 0.92. The model performed well in ordinary surveillance circumstances, but its detection efficacy decreased in cases of severe occlusions and low-light surroundings, as seen by a lower mAP@50-95 score of 0.512. YOLOv8 is compared with YOLOv7, RetinaFace, and Faster R-CNN, based on performance metrics. Given these challenges, YOLOv8 provides a solution for real-time face identification in security applications. Future research will focus on enhancing the model's resilience through sophisticated pre-processing approaches, multimodal detection, and transfer learning to improve performance in increasingly complex real-world scenarios.