Accident Detection using DenseNet

Abstract

According to a data, around 1.5 lakh persons die due to road accidents per year in India alone. 30-40 percent of these road accidents go unnoticed or neglected by the general public to avoid the unwanted police inquiry that can cost lives and time of several people. A simple idea can ease the process of controlling the traffic system and detecting accidents. The main goal of this work is to use computer vision and also deep learning to detect accidents through surveillance and dashboard cameras and then report it to nearby emergency services with valid accident images. So suppose we have L number of layers in a typical Dense Net structure there will be about (L*(L+1))/2 layers, so when n number of images are added it is simpler to process, because of the extended layers. Every layer adds only a limited number of parameters. This increases the flow of gradient through the network. Through this the task of DenseNet is accomplished.