Disasters caused by fire are considered among the most devastating, as they can contribute to climate change and result in loss of life, property destruction, and environmental and economic havoc. Early fire detection is crucial to prevent these losses and disruptions. Computer vision engineers have focused extensively on fire detection through vision detectors, recognizing it as a promising field of study. While low-level color characteristics were traditionally used for fire detection, deep learning models have significantly improved accuracy and effectiveness. This study proposes an approach for real-time fire detection in smart cities based on the YOLOv8 method, capitalizing on deep learning capabilities. The suggested smart city framework comprises four layers: application, fog, cloud, and IoT. By leveraging fog, cloud computing, and IoT for real-time data collection, potential harm is minimized. However, the IoT layer and internet connectivity introduce security risks. To address this, we integrate the secure surveillance method to mitigate these potential risks. Our method offers a quick, probabilistic, and lightweight approach to encrypt keyframes before transmission, considering the limited memory and processing capacity of devices, making it particularly suitable for IoT applications.