An Investigation in the Use of Inductive Loop Signatures for Vehicle Classification

Problem

If we could accurately sort the vehicles that use our highways into predefined classes, we could improve vehicle reidentification algorithms and traffic simulation models, gaining useful information to benefit highway maintenance, vehicle emissions management, highway design, traffic safety, and automatic toll collection systems. If we could do this using the current inductive loop technology already in place, we could gain all these benefits immediately, and with simpler equipment than is used by other commercially available systems.

Method

Using different pattern recognition techniques, we tested vehicle classification algorithms on the SR-24 freeway. We used existing loop detectors that can output a vehicle inductive signature. We divided vehicles into seven classes (car, SUV/pickup, van, limousine, bus, two-axle trucks, and larger trucks) to test two classification methods. The first method, a heuristic discriminant algorithm, used multi-objective optimization for training the heuristic algorithm. Some advantages of using heuristic algorithms include smaller data transmission and storage requirements, and a sequential discrimination procedure in which each stage can be individually fine-tuned. We developed three different heuristic algorithms which yielded 81 to 91% accuracy.

The second method used self-organizing feature maps (SOFM), which are artificial neural networks. A large advantage of using SOFM is that it needs only a small training data set to produce similarly high rates--more than 80% accuracy.

We also developed a speed measurement system using single loops, since it's necessary to accurately estimate speed in order to be able to judge a vehicle's size. Our method produced more accurate results than previous single loop estimation methods.

Different technologies work best with classification schemes suited to their particular method of signal detection; for this reason, using a combination of technologies will probably produce even more consistently accurate results.