A Hybrid RNN-CNN Approach to Automate Traffic Pattern Analysis in Roundabouts

Roundabouts have become a fundamental component in our urban transportation networks. Traffic pattern analysis is a critical step in the optimization of the design and utilization of those roundabouts. The traditional analysis performed by humans is labor intensive and cost prohibitive. In this work, we propose an alternative system that can automatically process a recorded aerial video of a roundabout and report the traffic patterns in it. The proposed system uses a pre-trained YOLO model to identify vehicles in the video frames. A hybrid RNN-CNN system tracks each vehicle from one frame to another as they travel through roundabout. Finally, statistical analysis is used to automate the counting of vehicles entering and exiting different sides of the roundabout. The output of the proposed system is a spreadsheet with the number of vehicles that entered/exited at all entry/exit points of the roundabout. The results show that the proposed system has an accuracy of 98.7<sup xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">%</sup>.