Vehicle re-identification with a single magnetic sensor

Vehicle re-identification enables the calculation of travel times and origin-destination matrices. These are key data for modeling and therefore optimizing the traffic flow. This paper compares the performances of different vehicle re-identification methods using the vehicle tridimensional magnetic signature recorded with a single three-axis magnetic sensor. These methods based on temporal distances and on features extraction are evaluated. The results obtained on a large data-base composed of signatures from 25 different vehicles show that the tridimensional Dynamic Time Warping is able to correctly re-identify 80% of the vehicles without any false alarms. To reduce its computational time, it can be combined with a Gaussian maximum likelihood classifier using temporal features through cascading rules without significantly reducing the re-identification performances. Moreover, the influence of the vehicle orientation to the North on the re-identification performances is evaluated and shown to be limited, which enables the re-identification of vehicles after a change of direction.