Vehicle Type Recognition In Traffic Video Surveillance

Intelligent transportation system is the trend of traffic development in the future, whose most important part is intelligent control. Based on traffic video monitoring, vehicle recognition can automatically identify vehicle type information by analyzing video data. Identification results can be used for automatic road toll, accessing control, traffic path planning and other applications. It has a broad market prospect and research value.The focus of the study is to design a complete vehicle recognition scheme through the experimental method based on traffic surveillance video, and take identification of buses and heavy tracks as an example. The research mainly consists of two parts:the first part is moving vehicle detection. In order to adapt complex road condition and real time application with moving shadow removal, each module of moving vehicle detection is well prepared and researched. Double Level background subtraction is used to get stable background image and rough moving region. To remove vehicle’s self shadow, a new method is proposed combining HSV color space and SILTP texture features to obtain more accurate moving vehicle region. Random noises from environment and small undetected regions are inevitable in obtained region, to solve this problem and get accurate and clear moving vehicle region, energy density analysis is adopted. The second part is the recognition of vehicle type, choose buses and heavy trucks as aim objects, whose size difference is small. Using a moving vehicle detection method from a large number of road environment videos to extract and store aim image files, and establishing a buses and heavy trucks sample database. Through analysis of the characteristics of vehicles, using LBP texture features based on HIK method and SVM off-line learning to get the vehicle type classifier, and then the classifier is applied to the real-time recognition.At the same time, in order to put the vehicle recognition scheme applied to the scene, develop vehicle type identification software system. The software on Windows platform using Visual C++MFC development environment, OpenCV image processing library, GSL scientific computing library and tools so on, it use for testing and displaying results of each vehicle type identification method module, such as background image, vehicle motion region, shadow suppression, filtering, vehicle recognition, can get the whole vehicle recognition scheme performance validation.Vehicle recognition scheme in traffic video surveillance is based on a large number of experimental results, and designed a software to test the on-site application. the proposed method can achieve a good vehicle recognition effect, and has a good real time performance.