Traffic Flow Measurement At Crossing Junction Based On Machine Vision

Intelligent Transportation Systems (ITS) are the coming urban traffic management system's main research directions. With the comparison of traditional traffic management method, intelligent transportation system monitors, schedules and controls traffic within a city or between different cities in order to achieve a larger area of traffic management. Intelligent transportation system consists of a traffic information collection system, traffic information processing and analyzing systems, and information dissemination system. Among them, the collection and procession of traffic information are a key system component, which is the important basis of intelligent transportation system to determine an accurate analysis. In the recent decade development process, the traffic information collection has mainly a sense coil, ultrasonic, radar, microwave, infrared and video capture mode. In kinds of acquisition mode of traffic information, the video capture mode is able to reproduce the scene of road traffic information and record information of each vehicles movement of the scene, so it has been widely used in practice.The paper targets traffic video as study object, design the number of movement vehicle movement detection algorithm. Specific research as follows:First, introduces the common detection algorithm of motion goal, including inter-frame difference method, optical flow method, and background subtraction method. And it analyzes the advantages and disadvantages of each method in the detection of vehicles.Then, when detects moving vehicles, in order to extract the background image of the traffic video image, sub-sorts prospect information, proposes that using a Gaussian mixture modeling method for vehicle detection. The paper observes mean and variance of each pixel, and establishes three background models, and determines each pixel of collecting video images. If pixel meets the model and put it into the model, if no model meets, round the smallest weight of the model and re-establish the model; pixels which do not meet the background as prospect information. As the foreground image contains noise and vehicle information which is lost, so filter the foreground image and mathematical morphology operations achieve the foreground information image separation of the background information image.When counts the vehicles, fills each vehicle the same data according to the vehicle's internal characteristics, different vehicles fills number followed by plus one, statistics the number of vehicles of each frame.Finally, using Matlab software simulates algorithm, experiments show that the algorithm can detect the number of moving vehicles at different speeds and statistics the number of vehicles. In addition, the separation method of the background image and foreground image is adaptive, and can be adapted to vehicle detection in different lighting conditions with good robustness.