Estimating time of arrival of vehicle fleets with GCN based traffic prediction

This paper presents an effective framework for estimating time of arrival of vehicles (buses) in an Intelligent Transit Management System (ITMS) having sparse position updates. Our contributions towards this is firstly in implementing a constrained optimization based road linestring segmenting framework ensuring ideal segment lengths and segments with sufficient density of vehicle position measurements which will result in valid statistics for scenarios involving sparse position measurements. Over this we propose a comprehensive approach for predicting traffic delays and estimated time of vehicle arrival addressing both the spatial and temporal dependencies of traffic. The traffic delay model is built on top of the T-GCN architecture on which we optimally augment an adjacency matrix which models a complexly connected road network considering the degree of influence between road segments, enabling the traffic delay model to look beyond physical road connectivity in predicting traffic delays and therefore producing better estimates of arrival times to points along the designated route of the vehicles.