Improving the accuracy of OD estimation from traffic counts employing a partial observabillity maximizing methodology

Traffic assignment models are essential tools in transport planning. One core input to these models is the Origin-Destination (OD) demand matrix. As this information cannot be observed directly, it is usually estimated indirectly, using techniques that retrieve information about the underlying traffic demand and its characteristics from traffic sensor data. It is therefore crucial that these sensors are strategically positioned in the network, herewith maximizing the amount of demand information that they can indirectly capture. In this paper we explore the impact on demand estimation quality of a new sensor location strategy, based upon the concept of (partial) observability in networks. We show that a novel strategy introduced recently by Viti et al. (2014), which exploits nothing more than the topological relationships between links, identifies sensor locations that enable OD estimation whose quality exceeds those of other partial observability approaches, as well as random locating strategies. An additional property of this approach is that it allows achieving performances close to methodologies based upon prior information, i.e. requiring data or extra assumptions in order to be solved, and that are developed specifically for flow estimation approaches (e.g., the Maximum Possible Relative Error metric of Yang and Zhou, 1998). We validate our approach empirically, through a mid-sized methodological case study and a practical instance, showing how our contribution allows attaining full OD coverage based solely upon link-route topological relationships, and how this property reflects in the quality of OD estimation in a practical instance, where full OD coverage cannot be guaranteed.status: accepte