Traffic Condition Assessment: Integrated Dynamic Travel Time Prediction and Accident Detection Models

Having access to accurate travel time is of great importance for both highway network users and traffic operators. The travel time which is currently reported for most highways is estimated by employing naïve methods and using limited sources of data. This might result in inaccurate travel time prediction and could impose difficulties on travelers. Therefore, proposing a comprehensive framework that utilizes various data sources and appropriate methodologies in order to predict the travel time accurately is essential. In this thesis, a hybrid dynamic approach is proposed to predict travel time of highway corridors under two traffic conditions: regular traffic condition and accident condition. To this end, first enhanced travel time prediction models for highway links are developed. The data, which is used in these models includes traffic spatiotemporal, geometry, weather condition, road work, special events, and accidents data. On the other hand, accident as an influential factor can considerably impact travel time. Therefore, advanced techniques including machine learning, deep learning, and deep ensemble models are developed to detect traffic accidents of highways in real time. Thereafter, a dynamic travel time prediction approach is suggested for highway corridors using integration of the developed travel time prediction and accident detection models of highway links. Finally, to find the optimal estimation of travel time for highway corridors, an ensemble based Kalman Filter algorithm is proposed that consequently led to boosting the prediction accuracy.