A Computer Vision-based algorithm for accident detection, severity classification and rapid emergency response using YOLOv8 and Twilio Programmable Messaging API

This paper introduces an algorithm designed to leverage the power of computer vision and deep learning for efficient accident detection, severity classification, and rapid emergency response. With the escalating road accident rates in India and a pressing need for timely medical assistance, our system addresses the critical gaps in the current emergency response infrastructure. The core of our proposed system lies in a sophisticated algorithm that processes traffic video footage using computer vision techniques. This algorithm not only detects accidents but also classifies their severity, providing a nuanced understanding of the gravity of the situation. In case of a severe accident, our system employs the Twilio Programmable Messaging API, through which emergency notification messages are sent to pre-registered hospitals, equipped with the precise coordinates of the accident site, enabling swift dispatch of medical assistance. The proposed system displayed a mAP50 of 97.4% during training and 97.1% during validation. Accident detection accuracy during testing stood at $98.56 \%$, while accident severity classification accuracy stood at $98.25 \%$. Emergency responses were recorded to have triggered in less than 5 seconds after the occurrence of a severe accident.