VARVO: a Novel Method for the Fast Detection of Vehicle Crash Events from Video Only Data

Abstract

Around 1,35 million people worldwide die each year because of traffic incidents, and it is estimated that another 50 million suffers serious injuries. This picture is particularly dramatic in the Andean Region where the death toll due to traffic accidents is as high as 127 deaths per million inhabitants. In recent years the deployment of the so-called Intelligent Transport Systems (ITS) across several developed countries has helped to reduce the number of deaths due to traffic accidents. An integral part of an ITS is the automatic detection of traffic incidents from video and sensor data. However, the scarcity of curated datasets, especially those that contained a reasonable number of positive instances of traffic incidents is hampering the development of artificial intelligence applications for the domain of traffic research. Given this scenario, we pursued answering the following research question: is it possible to detect car crashes through supervised machine learning based on the estimated speeds of cars from video only-data? Here we present VARVO, a novel algorithm for the detection of traffic incidents that does not rely on sensors for cars speed detection. VARVO performs a supervised classification task based on the sequential use of convolutional network-based object detection and bi-directional tracking. We also describe how the models implemented in VARVO improved their classification accuracy by applying an oversampling algorithm to deal with class imbalance. We believe that the deployment of VARVO could be linked to static traffic video cameras and could be part of the Intelligent Transport Systems foundations in other Andean countries.