IEEE TRANSACTIONS ON INTELLIGENT TRANSPORTATION SYSTEMS, VOL. 12, NO. 1, MARCH 2011 291 Requiem for Freeway Travel Time Estimation Methods Based on Blind Speed Interpolations Between Point Measurements Francesc Soriguera and Francesc Robusté Abstract—Travel time estimation from loop measurements has attracted extensive research in the last decade, resulting in nu- merous methodologies. Among these, those that rely on spot speed measurements at detector sites to obtain travel time estimation on the target stretch are the most intuitive. The key issue concerning these methods is the spatial generalization of point measurements over a freeway link. This paper shows that all speed interpolation methods that omit traffic dynamics and queue evolution do not contribute to better travel time estimations. All methods are in- accurate in congested and transition conditions, and the claimed relative benefits using various speed interpolation methods re- sult from context-specific experiments. Therefore, these methods should be carefully used and not taken as perfect. Lacking a better approach, it is recommended to avoid overcomplicated mathemat- ical interpolations and focus efforts on intelligent smoothing of the noisy loop detector data, reducing the fluctuations of short time interval aggregations while maintaining the immediacy of the measurements. Index Terms—Loop detector data, speed trajectory interpola- tion, travel time estimation. I. I NTRODUCTION AND BACKGROUND T HE ACCURACY of real-time travel time information disseminated on metropolitan freeways is one of the key issues in the development of advanced traveler information systems. Very accurate estimations could be obtained if suitable and intensive monitoring systems were widely available (e.g., direct travel time measurements by means of Global Position- ing System vehicle tracking or automatic vehicle identification (AVI) technologies such as video recognition of license plates or identification of the electromagnetic signatures of vehicles) [1]. Usually, however, travel time estimations must still rely on data obtained from the preexisting surveillance equipment that is most widely installed on freeways: loop detectors. The most intuitive and simple category of methods to obtain travel time estimates from loop point measurements uses the spot speed measurement at the detector site to generalize the speed over the target section and obtain the travel times. As Manuscript received April 15, 2010; revised July 5, 2010, September 2, 2010, and October 8, 2010; accepted October 26, 2010. Date of publication December 17, 2010; date of current version March 3, 2011. This work was supported in part by the Spanish Ministry of Science and Innovation under Grant TRA 2008-06470/MODAL. The Associate Editor for this paper was M. Brackstone. The authors are with the Center for Innovation in Transport, Technical University of Catalonia, 08034 Barcelona, Spain (e-mail: francesc.soriguera@ upc.edu; f.robuste@upc.edu). Digital Object Identifier 10.1109/TITS.2010.2095007 a result of this apparent simplicity and straightforward ap- plication using common aggregations of loop detector data, spot-speed-based travel time estimation methods have been implemented worldwide by most traffic agencies. Problems with these types of methods arise mainly due to two factors: 1) accuracy of spot speed measurements and 2) spatial gen- eralization of point measurements. The lack of accuracy of single-loop speed estimations, usually assuming constant ve- hicle length, is widely accepted and has been vastly analyzed for a long time [2]. Several methods have been proposed to enhance single-loop performance in terms of speed measure- ments [3], [4]; however, to solve this problem, most highway administrations have chosen to install double-loop detectors that are capable of accurate speed measurement at least on their metropolitan freeways, where intensive traffic monitoring is crucial. The spatial generalization of point measurements, which is necessary in this type of travel time estimation method, is the second factor that introduces error. The common practice is to assign a particular loop detector to a freeway portion and assume that the speed remains constant during the whole sec- tion and during the whole time aggregation period. Considering common loop detector spacing (i.e., 0.5 km at best), one can see that, in the case of a dramatic traffic state transition on the free- way section (e.g., change from free-flowing traffic to queued traffic within the section), the punctual speed measurements assumed for the whole section would be totally unrealistic, and the errors in the travel time estimation would be huge. These will be largest when most of the segment is queued and the detector is unqueued (simultaneously) or vice versa. This evidence leads to the obvious and widely demonstrated fact [5]–[7] that travel time estimation methods based on spot speed measurements perform well under free-flow conditions (this implies that there is not any change of a traffic state within the freeway section), whereas the accuracy of the estimation in congested or transition conditions are dubious. (There exists the possibility of a traffic state change within the freeway section and stop&go instabilities, which are not reported in the loop detectors measurements.) To solve this problem, several authors [8]–[12] propose new speed interpolation models, which are different than the con- stant assumption, to better describe the spatial speed variations between point measurements, particularly under congested con- ditions. Apart from a pair of remarkable exceptions and possi- ble alternatives in [11] and [12], where the classical continuum 1524-9050/$26.00 © 2010 IEEE