01042649

Component

This paper presents a new approach to the vehicle re-identification problem in the context of matching commercial vehicles crossing two adjacent weigh-in-motion stations at a commercial vehicle enforcement weigh station. A Bayesian formulation that utilizes finite mixture models to represent probability distributions of the attribute data is developed. Multiple combinations of vehicle attribute data (e.g., axle spacing, vehicle length, axle weights) are analyzed to represent different data collection scenarios (e.g., weigh-in-motion or vehicle classification only). The Bayesian model matches individual vehicles based on the posterior probabilities calculated from the vehicle attribute data and time stamps observed at the two stations. The results of the proposed method are promising with a mismatch error rate of 3% on a test dataset with more than 900 samples. For comparison purposes, another matching algorithm based on the nearest squared-distance is also developed. In all cases, the proposed Bayesian method outperforms this nearest neighbor method. The specific application that provides the framework for this paper is tracking individual vehicles through a weigh station system in order to improve enforcement efficiency and effectiveness. The paper discusses other potential applications of the re-identification algorithm and the associated future research.

01042056

07-2250

English

17p

2007

Transportation Research Board 86th Annual Meeting

CD-ROM

Figures (4) ; References (26) ; Tables (3)

Algorithms; Automatic vehicle classification; Axles; Commercial vehicles; Data collection; Identification systems; Vehicle classification; Vehicle length; Weigh in motion; Weigh stations; Weight

Bayesian approach; Reidentification

Highways; Terminals and Facilities; Vehicles and Equipment; I90: Vehicles

Transportation Research Board Annual Meeting 2007 Paper #07-2250

TRIS, TRB

Feb 8 2007 6:55PM