Research Of Vehicle Classification System Based On Multisensor Data Fusion

In recent years, the vehicle ownership surged, which leads to rapid growth of car traffic volume. Efficient and intelligent traffic management measures are required to resolve the traffic problems.In the background, as an important part of intelligent transportation system, vehicle classification system should play a bigger and more prominent role.This article summarized most of common vehicle classification systems, and analyzed their advantages and defects from the angle of application.On this basis, through the comprehensive comparison and analysis, a vehicle classification system which based on the fusion of laser ranging sensor and toroidal coil sensor is put forward.In terms of classification principle, laser ranging sensor collects vehicle body geometry data, toroidal coil sensor acquisition vehicle chassis data. Each type of data shows different kind of characteristics, and the fusion data couled improve the detection accuracy effectively. In terms of practical applications, this sensors combination can emphsis advantages of both single point laser sensor and toroidal coil sensor like small size, long life, easy maintenance, which shows a good application prospect.According to the beneficial effects above, this article made a deep research for vehicle classification which based on the fusion of laser ranging sensor and toroidal coil sensor, basing on practice. First of all, studied the overall design of the system, has been clear about the criteria for the classification of models, design a set of feasible scheme of vehicle recognition, and analyzes on the advantages of the scheme. Secondly, in this paper, SICK DT35 laser range sensor, single line loop detector and STM32 microcontroller were chosen to create a data acquisition hardware platform after an abundant research, and the laser sensor installation is specially designed. All of these devices have been set up in real freeway environment, and first-hand vehicles data has been collected successfully. Finally, based on the freeway vehicle data, 11 kinds of features were extrated. Through the fusion of those featrues, accuracy of vehicle classification has achieved above 96% by the random forest classifier, and the different results of fused data and unfused data were compared.Practice has proved that, double sensor fused vehicle classification system has a higher identification accuracy compared with the single sensor one. And compared with other types of vehicle classification system, the form of sensor combination in this paper has obvious advantages in the device size, service life and maintenance difficulty etc.