Optimizing Fire and Smoke Detection in CCTV Systems using Real-Time Detection Transformers

Fire and smoke detection systems are vital for public safety, enabling early warning mechanisms to prevent disasters. Traditional detection methods are often limited to confined spaces, necessitating the use of video surveillance systems for extensive monitoring. This study investigates the efficacy of various deep learning models, including Real-Time Detection Transformers (RT-DETR), for fire and smoke detection in diverse environments. By addressing challenges related to dataset quality and annotation accuracy, we demonstrate that RT-DETR significantly outperforms other models in both speed and accuracy, making it suitable for real-time applications. We also propose enhanced data preprocessing and augmentation techniques, such as histogram equalization, to improve model generalization. Our findings underscore the importance of robust data management in developing reliable detection systems, ultimately contributing to enhanced safety measures in smart cities and public spaces.