A Video-based Traffic Information Collection System For Multiple Vehicle Types

Traffic information of multiple vehicle types not only provides more comprehensivereferences for traffic control and management, but also promotes the researches of classictraffic flow theory and traffic simulation models. However most of video-based detectionsystems do not take multiple vehicle types into consideration nowadays, so a new video-basedtraffic information collection system for multiple vehicle types is developed. This imageprocessing-based system is developed by using OpenCV tool in the compiling environment ofVisual Studio2005. By tracking and classifying every passing vehicle under mixed trafficconditions, the type and speed of every passing vehicle are recognized. Finally, the flows andmean speeds of multiple vehicle types are output. The main processes of this system includedetecting the initial background image, calibrating traffic scene, setting road detection region,vehicle detection, shadow removal, vehicle tracking for speed, vehicle classification, datafusion and so on. Color image-based adaptive background subtraction is used to obtain moreaccurate vehicle objects, and a series of processes like shadow removal and setting roaddetection region are used to improve the system robustness. In order to improve the accuracyof vehicle counting, the cross-lane vehicles are detected and repeated counting for one vehicleis avoided. In order to reduce the classification errors, the space ratio of the blob and datafusion are used to reduce the classification errors caused by vehicle occlusions. And theK-means clustering is used to get more precise thresholds for vehicle classification. Thissystem was tested under four different weather conditions. The accuracy of vehicle countingwas97.4%and the error of vehicle classification was8.3%. The correlation coefficient ofspeeds detected by this system and radar gun was0.898and the mean absolute error of speeddetection by this system was only2.3km/h. This indicates that this system is reliable forcollecting traffic information of multiple vehicle types.