Inferring Network Origin-Destination Matrices Using Heterogeneous Traffic Sensor Information: An Optimal Sensor Location Policy

A trip Origin-Destination (O-D) matrix in a vehicular network is one of the critical components for transportation planning and/or traffic management. Because of the rapid development of intelligent transportation systems (ITS), trip O-D matrices could be directly obtained or indirectly estimated in light of the specific traffic information provided by various advanced sensor technologies; avoiding the problems associated with traditional O-D data survey approaches. Using heterogeneous traffic data sources collected by advanced sensor technologies, either passive-type vector detector or active-type automatic vehicle identification sensor for network O-D matrix estimation becomes feasible and cost-effective. However, due to a budgetary constraint of highway agencies, it is difficult to deploy various types of sensors in a full scale; an optimal sensor deployment policy in terms of the minimum number required and installation locations for an accurate O-D matrix estimate is another key element to be investigated. In the present research, the optimal sensor deployment problem for both passive and active sensors is respectively modeled and solved by a linear algebra-based method and a nonlinear programming approach, and the network O-D demand estimation problem is formulated as a path-based flow estimation model and solved by the pseudo-inverse matrix algorithm. Numerical analysis based on both hypothetical and simplified real networks is conducted to demonstrate the performance of the proposed model framework. Results of the empirical study indicate that the proposed path-based flow estimation model provides good estimates of O-D demands and path flows under different test network structures. In addition, a sensor deployment plan is optimally determined for the aim of obtaining a desirable O-D matrix estimate under a limited budget of the highway agency.