Journal of Advancement in Big Data Science and Data Analysis

Road traffic accidents cause millions of fatalities each year, and a substantial proportion of these deaths result not from the collision itself but from the delay between the incident and the arrival of emergency services. Conventional reporting mechanisms rely on bystander intervention or passive surveillance infrastructure, both of which introduce unacceptable response latency. This paper presents the design, implementation, and experimental evaluation of a Dashcam-Driven Real-Time Road Accident Detection and Automated Emergency Notification System. A dashcam mounted on the primary vehicle continuously captures live road footage, which is processed in real time by a YOLOv8 deep learning object detection model to classify surrounding road entities, including cars, motorcycles, trucks, and pedestrians. A deterministic, rule-based accident confirmation engine then evaluates inter-object bounding box Intersection over Union (IoU), trajectory deviation, and sudden velocity change across consecutive frames to distinguish genuine collision events from false positives. Upon accident confirmation, the system immediately queries a GPS module for precise vehicle coordinates and dispatches a structured alert email to designated emergency authorities via SMTP over Gmail, completing the entire pipeline within an average of 2.4 seconds. The system was validated in a controlled, two-vehicle experimental environment simulating five distinct collision scenarios. Results demonstrated a mean object detection confidence of 88.6%, an accident classification accuracy of 90.8%, a false positive rate of 4.3%, and a 100% alert delivery success rate. The prototype was developed at an estimated cost of Rs. 4,200 to Rs. 7,500, making it an economically viable solution for intelligent transportation systems, commercial fleet management, and smart city road safety infrastructure.