Traffic Nowcasting using Deep Learning

Nowcasting is the prediction of the present and the very near future of an indicator. Traffic Nowcasting is the prediction of various traffic factors occurring in the near future. This paper describes our approach, to predict short term city-wide high resolution traffic states with the static and dynamic information provided. We achieve this by utilizing the U-Net architecture to build a deep CNN model; test it on different cities to evaluate accuracies; and average the results at the end. With this, the model is better trained and will return more accurate, generalized results for different cities. The models are trained on traffic datasets provided by Traffic4Cast 2021 challenge. Thus, the aim of this project is to build a system for predicting and calculating traffic flow volume, heading and speed on a high-resolution whole city map by predicting the traffic states in the near future. The system could be utilized in a Traffic Prediction System and its scope entirely lies within traffic prediction.