Traffic State Estimation Using Speed Profiles and Convolutional Neural Networks

Abstract

Determining the traffic state is one of the most attractive problems for experts in the field of Intelligent Transport Systems (ITS). In this paper, a deep learning model for determining the traffic state is presented. Model is based on Convolutional Neural Networks (CNN) and uses weekly speed profiles as input data. The proposed model consists of input and output layer with an addition to four convolutional layers, two pooling layers and two fully connected layers that are extracting important features and classifying intersections as congested or not congested. We analyze data and predict traffic state for the most relevant road segments in the City of Zagreb which is the capital and largest city in Croatia. Speed profiles from included road segments are represented as one traffic image and used to train CNN. In that way traffic state for all sequentially connected road segments is estimated. The proposed method achieves a classification accuracy of more than 90% on three analyzed types of road topologies. The results show that CNN trained with traffic images can be used as a tool for traffic state estimation.