Research On SVM-based Vehicle Classification Method Using Geomagnetic Vehicle Detector

With the development of urbanization, the growing popularity of private cars, the road transport industry developing very fast, so the research of Intelligent Transportation Systems has been paid great attention. As an important part of ITS, Automatic Vehicle Classification (AVC) provides a lot of useful applications, such as transportation planning, road network designing and traffic managements. Vehicle detector with AMR Geomagnetic sensor is the study object of this paper, the advantages of this kind of detector are smaller size, lower cost, higher sensitivity, easier installation and maintenance.Firstly, the principle of vehicle detector with AMR geomagnetic sensor has been discussed in this paper, the vehicle classification detector is improved based on the previous two-node vehicle classification, vehicle classification detector designed in this paper uses a3-axis AMR sensor within a single node, the direction of the3-axis is orthogonal and respectively corresponding with the direction of height, length and width of vehicle, this kind of design not only enrich the collection of geomagnetic information, but also weaken the relationship between velocity and vehicle classification, because it could considering the relationship between vehicle classification and the entire vehicle shape but length, so it made single-node Geomagnetic vehicle classification detector possible.Secondly, we analyzing the characteristic of magnetic field signal collected by AMR sensor and using dynamic background value approach to separate the vehicle geomagnetic disturbance signal from background geomagnetic field, then abstracting the feature of vehicle geomagnetic disturbance signal and optimizing the feature set with Filter-Filter-Wrapper hybrid pattern feature optimization approach.Thirdly, after researching and comparing various kinds of multi-class SVM classification, we chose Directed Acyclic Graph Support Vector Machine (DAG-SVM) as the classification algorithm of AVC, we also improve the traditional DAG-SVM and proved it has smaller classification error in theory.Lastly, in order to verify the validity of vehicle classification algorithm proposed in this paper, by increasing video collection we building a vehicle classification verification system and conducting some field experiments at Beijing Jiaotong University and some streets around the campus. Field experiments show that the single-node geomagnetic vehicle classification detector proposed in this paper has good performance in classifying vehicle, the classification rate is very high, and achieved the expected goals.