Residual Neural Network–Based Smart Accident Management System; Running head: ResNet-Based Accident Management

Abstract

The primary cause of accidents in recent years has been driver fatigue, which has resulted in serious injuries, fatalities, and financial losses. This paper proposes a smart accident management system that combines various technologies to enhance road safety significantly. Three features are implemented in this work. First, a mechanism is developed in such a way that if the driver is found to be an alcoholic, it is sensed using the MQ3 sensor; then the engine will not start. Second, in case the driver is drowsy or yawning, it is detected using eye aspect ratio and mouth aspect ratio, which are calculated using a live video feed of the driver while driving the vehicle. The facial features or patterns of eye and mouth movements that are associated with drowsiness and yawning are detected by using a residual neural network classifier. The results are experimentally validated and could successfully detect drowsiness/yawning. If the driver is found to be drowsy or yawning, a wake-up message is generated to alert the driver. Third, in case of accidents, it will be sensed by a vibration sensor; intimation will be sent to the concerned people using GPS (Global Positioning System) and GSM (Global System for Mobile Communications) modules. The concerned people get to know the exact location so that they can rush to the place. Overall integration of these features will help in minimizing and managing road accidents. In most of the existing works, data available is used, whereas in this work, an own data set is prepared from real-time data. With this real-time data, the model accuracy is found to be 98%, which is better than the existing results. The novelty of this work is that the driver’s state is continuously monitored; using a live video feed, drowsiness is detected, and simultaneously, alcohol consumption is also detected. The alarm signals are generated even if one condition is true, and all three safety features are integrated as a single unit in the prototype developed.

Cite this article as: D. Ch, J. K, P. B, R. MR, H. P. Lee and G. Kasilingam, “Residual neural network–based smart accident management system,” Electrica, 26, 132, 2026, doi:10.5152/electrica.2026.25132.