Development of an Accident Detection System on The Highway Using the YOLOv8 Algorithm and Automatic Notification Via Telegram

Abstract

This research develops an accident detection system using supervised machine learning with YOLOv8 for object detection. The stages include data collection, labeling, model training, and system implementation with Telegram notifications. Data is taken from sample videos, converted into frame-by-frame images, and labeled with LabelImg. YOLOv8 is trained to recognize five object classes: car, accident, truck, person, and motorcycle. Implementation is done in Python with OpenCV, ultralytics, and cvzone. The system sends real-time notifications to Telegram upon an accident, achieving an average accuracy of 0.914 with notification times of 287.2ms – 334.1ms. This system aids traffic monitoring and quick response to accidents, reducing the negative impact of traffic accidents.