Comparing the accuracy of real-time transit arrival estimations with a Kalman Filter motion model in different transit networks models

Estimation of arrival time at transit stops is an active research field. Arrival time can be estimated with forecasting, stochastic, or learning methods. Regardless of the arrival-time estimation model, the transit lines must be modeled to identify the line segments where vehicles move. Hence, different network models can be created to identify when a transit vehicle is moving. The presented paper uses the Kalman Filter motion model to estimate transit arrivals on different network models. It is shown that the Kalman Filter motion model estimations vary significantly depending on the network model. The paper estimates arrival time in a real-life case study, using eight network models of transit lines: dwell, dwell/intersection, interval, stop, intersection barrier, interval/barrier, and adaptive based. The adaptive and interval/barrier-based network model leads to the most accurate results when the Kalman Filter motion model estimates arrivals.