Fire presents a significant hazard, capable of causing extensive harm to people, property, and the environment. Traditional fire detection systems, while useful, often lack reliability due to their dependency on sensor-based technologies for detecting fire and smoke. This paper proposed a novel ensemble algorithm designed for the accurate detection of fire and smoke for custom and CCTV footage dataset. The ensemble model includes two components: a highly accurate fire and smoke detection model and a specialized YOLOv8 (You Only Look Once) model for smoke detection. This integrated approach, combining the robust fire and smoke detection model with the enhanced capabilities of the YOLOv8, marks a significant advancement in real-time fire and smoke detection within CCTV surveillance systems. The models were trained using two Roboflow datasets. The first dataset contains 5000 images categorized into fire, nonfire, and smoke. The second dataset, focused on smoke, comprises 18,000 images, predominantly featuring 10,000 instances of smoke, with the rest labeled as null. Additionally, the dataset was extended by annotating videos from various online sources, resulting in 1290 additional images encompassing fire, smoke, and nonfire classes. After augmentation techniques were applied, the dataset expanded to approximately 3000 images. Our two models obtained a precision and recall of 0.98 and 0.99, respectively. The paper further discusses the development process, the algorithmic intricacies, and the practical implications of deploying this integrated system, underscoring its potential to revolutionize fire safety measures across various settings.