Real-Time Accident Detection and Intimation System Using Deep Neural Networks

Abstract

In India, around 4.61 Lakh road accidents happened in 2017 out of which 1.49 Lakh led to fatality. It is estimated that Andhra Pradesh alone takes a share of 7416 deaths among them. Among the total accidents, the death tolls are about 55,336 only from the two-wheeler crashes which indicate the pathetic and alarming scenario of road accidents in India. Many lives would have been saved in such conditions if the accident vehicle is detected, and the information regarding the incident is sent to the right people in right time. This situation motivated us to take up this research, which detects the road accidents using a computer vision system built around a Raspberry Pi and intimate the registered mobile numbers through IoT. The vehicle accident detection system (VADS) is built around a Raspberry Pi interfaced with a Web camera. The camera may be fixed in places like four road junctions, T-junctions, and other important locations where the probability of accident occurrence is high. The camera continuously captures the scene under consideration and gives the input to the processor. A convolution neural network architecture is designed and implemented to classify the severity of the accident into one among the three categories: good, moderate, and worst. As and when the test image in the scene is classified as a “moderate” class or a “Worst” class, the system identifies the situation as a serious condition and immediately triggers an event in the Ubidots cloud by using Wi-Fi interfaced with the controller. On triggered by the event received from the processor, the cloud immediately delivers a text message to the registered mobile numbers like: ambulance or a police control room to ensure immediate help that saves the life of the victim.