Object Detection in Foggy Conditions using Deep Neural Networks

Object detection in foggy environments is a hard task owing to the image quality degradation brought about by diminished visibility. Fog hides primary features, and it becomes challenging for regular object detection models to effectively identify and locate objects in real-time. Current approaches like Dark Channel Prior (DCP), Contrast Enhancement Prior (CEP), and other conventional deep learning models like YOLOX find it difficult to achieve high accuracy under such extreme conditions. In order to overcome these shortcomings, this work introduces a new method by combining a dehazing module, AOD-Net, with a Gradient Reversal Layer (GRL) to enhance domain adaptation between the synthetic and real foggy scenes. The model utilizes YOLOv8, a current state-of-the-art object detection method, which enhances real-time detection accuracy with its state-of-the-art architecture. This work contributes to the development of intelligent transportation systems, autonomous vehicles, and surveillance technology by providing an efficient and strong solution for detecting objects in real-time in low-visibility scenes. The Performance Metric are Mean Average Precision (mAP) , F1 Score and Recall to evaluate the model's accuracy in detecting and localizing objects in foggy environment in real-time conditions. In our proposed method not only improves detection accuracy on foggy scenes but also Reduces False Positives and Missed Detections, especially in the presence of Moving Objects.