Predictive neural networks and fuzzy data fusion for on-line and real-time vehicle detection

This article describes an on-line and real-time vehicle detection system. This system detects vehicles passing over magnetic sensors. It works independently of their initial position and of strong magnetic disturbance possibly induced by the load carried on the vehicles. This system is based on the co-operation between a reflective agent, using a reliability measure of its output, and a detection agent (on which this article mainly focus) based on two predictive neural networks and model selection techniques. The fusion of the data delivered by each agent is obtained through fuzzy logic rules. The system is also strengthened to resist substantial magnetic disturbances (even non-periodic ones); it uses the three components of the magnetic field, and is rotational invariant. Furthermore, its modular design opens up many possibilities of evolution.