This research proposes a robust real-time fire detection framework based on YOLOv11 trained on a large-scale and diversemulti-source dataset containing more than 105246 fire and non-fire images. The dataset was curated from multiple publicsources including Roboflow, D-Fire, Industrial Fire, forest fire datasets, and a custom dataset collected under differentenvironmental conditions. To improve generalization and reduce background misclassification, several augmentationtechniques including brightness adjustment, blur transformation, HSV modification, rotation, and noise injection wereapplied during training. The proposed model was trained using NVIDIA Tesla P100 GPU resources on Kaggle and optimizedto reduce false alarms caused by red, yellow, and orange background objects. Experimental evaluation demonstrated asignificant improvement in detection reliability, achieving approximately 77% detection accuracy with reduced backgroundfalse predictions compared to earlier baseline experiments. The trained system was further integrated with a Streamlit-baseddeployment interface for practical real-time testing. The proposed framework provides a scalable and cost-effective solutionfor intelligent fire surveillance applications in smart cities, industrial safety systems, and public monitoring environments.The achieved experimental results show that the proposed solution is suitable for creating a smart and real-time videosurveillance system for fire/smoke detection. Fire accidents represent one of the most dangerous threats to human life, industrial infrastructure, forest ecosystems, andpublic safety systems worldwide. Rapid fire propagation can cause irreversible damage within a very short duration, making early-stage fire detection extremely critical for minimizing casualties and economic losses. Conventional fire detection systems primarily rely on smoke sensors, thermal detectors, or gas-based alarm mechanisms. Although these systems are widely used, they often suffer from delayed response times because they can only trigger alarms after sufficient smoke accumulation or temperature rise has already occurred. Furthermore, traditional sensorbased systems may fail in large open environments, outdoor areas, industrial plants, or locations where smoke dispersion patterns are unpredictable. Recent advancements in Artificial Intelligence (AI), Computer Vision, and Deep Learning have significantly transformed automated surveillance technologies. Intelligent vision-based fire detection systems can analyze visual information from cameras in real time and identify fire hazards at an earlier stage compared to conventional methods. Among various deep learning approaches, object detection models based on Convolutional Neural Networks (CNNs) have demonstrated remarkable performance in detecting dynamic visual patterns such as flames and smoke. In particular, the YOLO (You Only Look Once) family of object detection algorithms has gained substantial attention because of its high detection speed, lightweight architecture, and suitability for real-time applications.
Kumar et al. (Wed,) studied this question.