Vehicle Detection And Classification Using Magnetic Sensors

The urban traffic congestion is getting more and more serious.It causes huge economic loss and it brings other problems related to traffic safety and environmental pollution.Intelligent transportation systems is a kind of informationalized and intelligent modern transportation system which can improve the level of traffic safety and the traffic efficiency.It can also promote the development of transportation industry,so as to alleviate traffic congestion.Vehicle detector is the tool of collecting traffic parameters such as traffic flow,vehicle speed and vehicle type.It is an essential part of intelligent transportation systems and it provides basic traffic data to intelligent transportation systems.Recently,vehicle detectors based on magnetic sensing technology have been widely concerned by its low cost,easy installation,high sensitivity and scalability.However,many disadvantages still exist in related researches such as difficult installation,blocking traffic and cannot rule out the interference from the adjacent lane.In order to overcome the deficiency of related researches,a kind of vehicle detection and classification system based on magnetic sensor is designed in this paper.First,the introduction of the principle of vehicle detection and the system platform are presented.After analyzation and feature extraction of vehicle magnetic signals,a single-lane vehicle detection algorithm is designed.Experimental results show that this algorithm can achieve high accuracy.In order to reduce the interference caused by vehicles driving in the adjacent lane,this paper proposes a preprocessing strategy based on threshold optimization.Then the Classification And Regression Tree(CART)algorithm is used to identify the driving lane of each vehicle.By this way,the interference can be effectively overcome.Finally,this paper analyzes and extracts the features of vehicle types.K-Nearest Neighbor(KNN)and Back-Propagation(BP)neural network algorithms are utilized for vehicle classification.Experimental results show that both algorithms have good performance.