CI/CD Pipeline for MLOps: Real-Time Fire Detection Using YOLO and Computer Vision

Abstract

Fire accidents pose severe threats to human life, industrial infrastructure, public safety, and environmental resources. Conventional fire detection mechanisms rely primarily on smoke sensors, heat detectors, and manual surveillance, which often result in delayed alerts and limited spatial coverage. This paper presents a Continuous Integration and Continuous Deployment (CI/CD)-enabled Machine Learning Operations (MLOps) framework for real-time fire detection using a custom-trained You Only Look Once (YOLOv8) object detection model integrated with computer vision techniques. MLOps practices are employed to automate the training, deployment, monitoring, and maintenance lifecycle of the detection model, while the CI/CD pipeline ensures continuous integration, automated testing, and rapid, consistent deployment of application updates. The proposed system captures live and prerecorded video streams, performs frame-level preprocessing using OpenCV, and applies the trained YOLOv8 model to identify fire instances with high accuracy and low inference latency. GitHub, Jenkins, Docker, and AWS EC2 are integrated to automate the complete machine learning deployment workflow: whenever new code is pushed to the GitHub repository, Jenkins automatically triggers the pipeline, builds a Docker image, deploys the containerized application, and starts the fire detection service on a cloud instance. When fire is detected above a predefined confidence threshold, the system generates audio alarms, dispatches email notifications, and logs detection events for subsequent monitoring and analysis. Experimental evaluation demonstrates that the proposed framework achieves high detection accuracy, low inference latency, and reliable, repeatable automated deployment. The results confirm that combining MLOps discipline with CI/CD automation substantially improves the scalability, reproducibility, maintainability, and operational efficiency of real-time computer vision applications.