Road Accident Detection using Deep Learning

TIt is estimated that 1.3 million people die each year from traffic accidents. The response time of paramedics plays an important role in determining survival probability. At the site of an accident, the response time variable depends on the actions of pedestrians and traffic officials. Using a CNN with alternate max-pooling layers, we propose using CCTV camera footage to classify images as accidents or non-accident. Using alternating max-pooling layers helps detect accidents even if the vehicle is not in the centre of the frame. We prepare our dataset from a variety of sources due to the lack of accident datasets in this field