Traffic Congestion Detection from Surveillance Videos using Deep Learning

Abstract

Countless cameras, both public and private, have been installed in recent years for the objectives of surveillance, the monitoring of anomalous human activities, and traffic surveillance. Numerous worrisome and aberrant actions, such as theft, aggression, and accidents, make it difficult to notice and recognise such behaviour in a real-world setting. The topic of this study is car wrecks as depicted in online videos of traffic. Modern traffic monitoring and surveillance rely heavily on video traffic surveillance cameras (VTSS). Consequences of a rapidly expanding human population include a higher frequency of accidental injuries. The VTSS is employed to identify unusual occurrences on various roads and highways, such as traffic congestion and car accidents. When accidents happen on lengthy roadways or in remote areas, victims are often powerless and some don't make it. The purpose of this study is to provide a method for automatically identifying incidents in surveillance footage. Convolutional-neural-networks (CNNs), a specific deep learning approach developed to cope with grid-like data, have been shown to be useful in image and video processing, according to a study of the relevant literature. This study use a rolling prediction method and convolutional neural networks (CNNs) to detect accidents in VTSS footage. A dataset of anomalous photographs, called the Vehicle Accident Image Dataset (VAID), was created and used in the training of the CNN model. The proposed method was put through its paces by analysing data gathered from running the trained CNN model on a number of different films. This study's findings demonstrate a 93% success rate in identifying traffic accident incidents in films from traffic surveillance systems.