Using Image Processing for Studying Microscopic Traffic Behavior

There are many methods developed for measuring the qualitative and quantitative parameters of traffic, but few research works have been done for measuring the position of vehicles. In these research works such as the one presented here, position of every vehicle is determined and according to the application, other traffic parameters can be calculated using the position of vehicles.For studying the microscopic movement behavior of vehicles in the basic freeway section, it is important to determine the position of vehicles in every time step. In this research a system has been developed which determines the position of vehicles in the considered section of freeway using simple image processing algorithms. Simplicity of the algorithms has reduced the execution time of the prepared image processing software and accuracy of the positions of vehicles is about the dimension of a private car, this accuracy is enough for most of the microscopic traffic studies. Input of the developed system is a video film of the vehicle movements and the output is a table of vehicles positions in the freeway.Using the table of vehicles positions in the freeway, trajectory of vehicles can be determined. Trajectory of vehicles can be used for modeling the drivers’ behavior. Driver's behavior models can be used for micro simulating of traffic. In addition, macroscopic traffic parameters including, number of passing vehicles, average speed, and other traffic characteristics can be calculated. Using image-processing techniques instead of manual methods increase the accuracy of measurements and a tedious work for human can be done by machine. Therefore, larger statistical samples can be collected. In addition, obtained data can be relied on and expenses are cut down. Regarding the fact that lack of safe distance observance is one of the main reasons of collisions, calculation of longitudinal distances between vehicles is introduced as an application of the developed system.In many images taken from the streets, in addition to the considered vehicles, there are other vehicles and pedestrians moving near the considered street. Thus, it becomes important to distinguish between the considered vehicles and other moving objects in the taken images. In this research, the considered street has been divided into windows and subsequently is used for image processing. The size of these windows is determined as if they can produce a complete view of the considered street with enough resolution and accuracy.In the developed system, position of vehicles can be determined without needing special devices. In addition, without knowing pan, tilt, focus, and height of the camera, only with knowing the co-ordinate of 4 points of the image in the real world, distances and positions in the street can be calculated. The basis of the developed image processing system is a projective model that a point on the street can be related to a respective point on the screen. Thus, all points on the street plane, which can be seen in the camera picture, are known. Windows are arranged in horizontal rows. Each row contains a number of windows. Considering the nonlinear projection of 3D images to 2D images, vertical distances between rows of windows are determined by a nonlinear equation. The distribution of the windows is determined as if all of the distances between rows of the windows show a specified distance, e.g. 1 meter, in the street. In this way each window would be matched to a specified space on the street. Each window is processed for detecting whether there is a vehicle in it or not. In the developed computer program, an effective algorithm has been used to detect the vehicles in individual windows. In this algorithm, averaging the pixels of that window in the sequence of all images, produces background of that window, and then moving vehicles can be detected by differentiating each image from its background. In the differed picture, pixels, which have high values, suggest presence of moving object, which in this case is a vehicle. When the number of pixels related to a vehicle exceeds a threshold, that window is considered to detect a vehicle. Such a process is done for all the windows in the images. In this way, all vehicles can be detected in the specified sector of the street and subsequently the position of vehicles can be determined and calculated in every image. If a vehicle can be seen through any of these windows, the corresponding value of that window will become 1, otherwise it will become 0. In heavy traffic when the images of vehicles conflicts and frontier edge of vehicles cannot be seen this system looses its efficiency.