Design and Development of a Real-Time Traffic Congestion Analysis System

The number of vehicles on the road has drastically increased over the years, and it has become the need of the hour to monitor and analyze traffic congestion in real-time. With modern technology, high-resolution closed-circuit television has been installed on several major junctions to monitor traffic violations. In this paper, an attempt is made to design a traffic congestion analysis system that uses a combination of cameras with the concepts of computer vision and machine learning to analyze realtime traffic congestion. The standard you look only once (YOLO) algorithm is considered for the proposed work, and the evaluation of the proposed system was carried out using videos collected from a few skywalks in Bangalore. Samples from Indian driving datasets with additional annotated images of vehicles from Google images were used for training and validation. The performance evaluation of the proposed real-time model was carried out on videos captured from a mobile phone and real-time input from a laptop camera. Recognition accuracies ranging between 76.70 – 89.24% and 83.08 – 92.65% are obtained using the proposed model with YOLO v3 and YOLO v4, respectively. The proposed system can generate analysis reports providing information on total traffic congestion in a particular stretch or junction with additional details, including the type of vehicle contributing to the total traffic congestion and the traffic congestion for an entire day.