Research On AMR-sensor-based Traffic Information Detection Algorithm

Presently, transportation problem like traffic congestion is one of the biggest problems which restricts the development of the city, and the Intelligent Transportation System(ITS) as an effective way to solve the traffic problems becomes increasingly prominent. How to acquire the accurate and real-time information of the road traffic states is the key to construction of ITS, which provides powerful data support for the transportation system. For this reason, a traffic information detection system based on AMR(Anisotropic Magneto Resistance) sensors is designed in the paper, a real-time vehicle detection algorithm through baseline tracking and an improved DAG-SVM(Directed Acyclic Graph Support Vector Machine) vehicle classification approach are proposed, which will effectively improve accuracy and efficiency of detection about the traffic flow, vehicle type, vehicle speed and lane occupancy rate. The paper mainly covers the follow areas:(1) A novel traffic information detection system based on AMR sensors is presented. The system consists of two parts: the information acquisition subsystem and the information processing subsystem. In the information acquisition subsystem, the AMR sensor nodes are used to detect the features change of the earth magnetic field which will be disturbed by passing vehicle. The Receiver is used to transmit the factures to the information processing subsystem. In the information processing subsystem, the traffic information will be acquired by analyzing the features.(2) A new Real-time vehicle detection algorithm through baseline tracking is proposed. By measuring the disturbance of the Earth magnetic field, weighting function is designed to correct base value in time, and state machine is used to detect vehicle. The Filter-Filter-Wrapper model is used to extract features which is better to vehicle classification through field experiment, and these extracted features will be the effective feature vectors which are used for vehicle classification.(3) According to the feature vectors, DAG-SVM is chosen as the method of Vehicle classification. After analyzing the error accumulation of the DAG-SVM, an improved DAGSVM vehicle classification approach was put forward. In this approach, the distance among two classes calculated by features with the feature weight was introduced to build directed acyclic graph.(4) In order to verify the validity of the traffic information detection methods proposed in this paper, field test system has been built, and the vehicle detection algorithm was tested on the spot. DAG-SVM classifier was trained and tested to classify the vehicle type with the sample set which was acquired in the experiment. Experiments have showed that the accuracy of the vehicle detection test was above 98%, while the accuracy of the vehicle classification test was more than 75%, and the improved DAG-SVM was better than the traditional DAGSVM.The AMR-Sensor-Based Traffic Information Detection Algorithm proposed in this paper can effectively obtain traffic information. It is of high popularization value and broad application prospect.