Microwave Radar And Video Sensor Fusion For Vehicle Detection And Classification Technology Research

With the rapid economic development, traffic demand and infrastructure are becoming more and more contradiction, so intelligent traffic system is design to solve this problem. The system is a new transport system which using advanced information technology, data communication and transmission technology, electronic control technology and computer processing technology. This technologies could maximize the efficiency of the transportation infrastructure.And this system has been get great attention and development in China.Vehicle detection and classification is a very important part in intelligent transportation system. It provides a wealth of traffic flow information, such as traffic density, vehicle speed, vehicle profiling, vehicle type and road traffic incidents. These are valuable information for applications of traffic signal control, road surveillance, road planning and so on. However, the existing vehicle detection and classification technologies are limited to a single sensor mode. The traditional single sensor for vehicle detection include coil detection, laser detection, infrared detection, video detection and radar detection and so on. The detection efficiency and classification results are subject to a lot of constraints. Some sensors are difficult for installation, some are vulnerable to the environment and weather, and some others provide the information too homogeneous. They are unable to meet the needs of development of intelligent transportation systems.In this paper, a system using microwave radar and video sensor fusion for vehicle detection and classification are proposed to solve the problems mentioned. We build the hardware platform for microwave radar vehicle detection and video vehicle detection. And in the platform we fusion the information which are collection by the two sensors for vehicle classification. We could obtain vehicle height profile from the microwave sensor and vehicle planar contour from the video sensor. After vehicles feature extraction gaussian mixture models (GMM) are used to establish the matching templates. Vehicle classification under the matching templates is based on bayesian network which plays the role of platform for integrating the different vehicle features into data fusion system. Using bayesian network to build the system could judge the different features scientifically and quantify the features evaluation. And the more advantage that the entire system is highly configurable and robust. Experimental results show that the proposed system and algorithms have very stable performance of vehicle detection and can improve the vehicle classification accuracy rate from 79% which only using microwave radar to 87% which using sensor fusion, and especially reduce the significant classification error rate from 9% down to 2% between the small and medium sized vehicles and large vehicles.Finally, the system using microwave radar and video sensor fusion for vehicle detection and classification is used on the road for testing. After the actual test shows that the system achieves the desired objective and is stable and reliable. This system has a bright future in intelligent transportation systems and traffic monitoring on multi-sensor fusion. We believe the system will greatly facilitate the research and development activities for intelligent transportation systems.