AI‑Powered Video Surveillance for Enhanced Object Detection and Incident Monitoring Using YOLOv8

Abstract - Timely and accurate detection of safety-critical incidents, such as vehicle accidents and human falls, is crucial for improving real-world surveillance-based emergency response. In this work, we investigate the use of YOLOv8-based object detection models, trained on small, domain-restricted datasets, for event detection. All variants of the model were deployed through a Flask-hosted web interface, allowing for extensive testing on static images as well as uploaded video streams. On a range of hardware platforms, the detectors showed well-balanced precision-recall performance with near real-time appropriateness of inference times. An added alerting module further enhances the pipeline, with an integrated solution for automated incident alerting. These results illustrate the viability of using light YOLOv8 models in efficient surveillance contexts in smart cities and industrial processes. Key Words: YOLOv8, video surveillance, object detection, incident detection, vehicle collision detection, human fall detection, Flask web interface, real time monitoring, alerting, deep learning.