Integrated Approach for Video Frame Dehazing and Living Being Detection for Fire Case Scenarios

This research presents a sophisticated deep learning-based framework designed to improve visibility and detect living beings in environments affected by fire. The proposed method operates in two key phases: first, the dehazing and desmoking of images and video frames, and second, the detection of living beings within these enhanced visuals. The initial phase leverages the Feature Fusion Attention Network (FFA-Net) to effectively eliminate haze and smoke, resulting in significantly clearer and more detailed imagery. The second phase employs the YOLOv8 object detection model, known for its precision and speed, to accurately identify and localize living beings within the processed images. This integrated approach ensures that the visual data is both clear and that the detection process is timely and accurate, which is crucial for coordinating rescue efforts in disaster scenarios. The proposed method has been rigorously tested across various datasets and environmental conditions, demonstrating superior performance in both image enhancement and detection accuracy, making it a reliable tool for improving situational awareness and supporting decision-making in fire rescue operations.