Generative Desmoking Algorithm for Enhancing Images in Low-Visibility Environments
摘要
In emergency scenarios, such as fires or industrial accidents, clear visibility is essential for making quick, effective decisions and ensuring the safety of both responders and individuals at risk. Limited visibility due to smoke and haze can hinder rescue operations, delay critical actions, and increase the potential for harm. To address this challenge, our dehazing solution is designed to enhance real-time video feeds by instantly reducing smoke interference, providing firefighters and emergency responders with vital visual clarity in smoke-obscured environments. This advanced dehazing technology leverages state-of-the-art GAN algorithms to process live video streams, clearing smoke in real-time and preserving essential details even in dense, low-visibility conditions with a SSIM Score of 0.87 and PSNR of 30.2. Unlike traditional dehazing or image enhancement methods that are either slow or not tailored for real-time use, our device/software operates with minimal latency, ensuring responders have an uninterrupted, clear view of the scene. Using NH-HAZE Dataset as base for training along with augmentation and masking techniques helped it achieve significant enhancement in situational awareness, enabling responders to make informed decisions and safely navigate hazardous environments. Moreover, the integration of human detection capabilities makes the device/software highly versatile and valuable across a range of emergency response fields. By identifying and highlighting human forms within smoke-filled spaces, the device assists in quickly locating individuals in need of rescue, even under challenging visual conditions. This dual functionality—combining dehazing with human detection—optimizes rescue operations, reduces search times, and helps direct responders to individuals who may be trapped or in immediate danger.