Vehicle Classification with low computation magnetic sensor

A vehicle classification algorithm is proposed based on signal obtained with low computation magnetic sensor. We focus on an implementation of our algorithm on wireless magnetic sensor node. Since the sensor node has limitations on computing capacity and power source, the algorithm must not be too complex. Features we choose to extract for classification include normalized vehicle magnetic length, averaged energy, and number of peaks in Hill pattern. These features are very simple to obtain. We classify vehicles into 5 types: motorcycles, cars, pickups, vans, and buses. The classification based on these features shows promising results. It can identify motorcycles and buses with very high accuracy. The group of cars, pickups and vans possesses similar distribution of features. The classification therefore must consider the three features in a tree structure form. A much better result obtained if we consider cars and pickups belonging to the same class.