Vehicle Recognition

To determine the average speed of a verhicle on a somewhat longer travel path it is necessary to capture its features at some first detection site and recognize it at some second one. The same problem arises when trying to trace the travel path of a vehicle, e.g. through a tunnel.

This requires some kind of description of the vehicle to be sent either from the first detection site to the second or from both of them to a third place. This description must be detailed enough to not confuse different verhicles, but also summarized enough to remain the same in spite of inevitable measuring incertainties. Also, the more detailed a description is, the more computing time it takes to perform an adequate comparison.

Magnetic signatures as acquired by an inductive loop are very significant "fingerprints", but comparing them by cross-correlation is very time-consuming. This is due to the fact that matching signatures must first be synchronized before cross-correlation can recognize them as matching. With no further means for synchronization this requires many cross-correlations with different delays to be computed; the amount of the maximum result of these computations is then a measure for the similarity of both signatures. Moreover, prior to being able to synchronize the signatures they must be compensated for speed. This requires speed measurements at both measuring sites, each one – when performed with inductive loops – requiring another computation of a whole sequence of cross-correlations.

The inductive loop of a CLASSAX sensor setup can easily capture such signatures. But now the axle sensors deliver not only accurate reference points for synchronization but also axle speeds for compensation, where even acceleration or braking can be taken into account. Furthermore, the same vehicle data which are used for AVC can here be used to perform a pre-selection: It is not necessary to check whether two vehicles with different axle spacings or even different numbers of axles might be the same one. Otherwise, the comparison requires only computation of one single cross-correlation because speed compensation and synchronization have already been achieved independently.

Thus, using CLASSAX vehicle recognition requires only a small fraction of the computational power otherwise necessary. Investing some of the savings in additional measuring sites is then a way to enhance the performance dramatically.