Accident Detection System

Road traffic accidents remain a major global public‑health challenge, where delays in detection and poor contextual information often increase the severity of outcomes. This paper presents ADS AI, a practical, end‑to‑end framework that combines a lightweight, low‑latency detection stage with a deeper, multi‑modal reasoning stage to deliver fast, actionable intelligence for emergency response. The first tier employs a YOLOv8 object‑detection model for near‑real‑time identification of anomalous events from live video streams; the second tier uses a multi‑modal generative model to create structured, human‑readable incident summaries and recommendations. The platform integrates persistent storage, geospatial analytics, and automated alerting via messaging APIs to form a deployable prototype. We describe the architecture, implementation choices, evaluation approach, and future directions, highlighting how this two‑tiered design enables both speed and situational depth—shifting traffic safety systems from reactive monitoring toward data‑driven, proactive operations.