Research On Detection For Traffic Violation Based On Video Technology

With the quick increase of vehicle quantity worldwide, the transportation status is thought much increasingly. How to carry on the transportation management availably becomes the focus that the international governments and the departments pay attention to. Aiming at this problem, various intelligent transportation management systems are created or under development, to monitor and control the cars drive in violation is most important tache of the systems. Traditionally, to monitor and control the illegal vehicle is mainly get across the underground sensor or axles machine to realize, these methods have damaged to the road, whose construction and installation are all inconvenient and usually need to be maintained and supported. Along with the high speed of video technology and flat-out times increases of the computer's speed, the realization of traffic monitoring system based on computer video detect technology becomes the possible. Compare with other traffic video detect technology, computer video detection technology has many advantages, the installation and adjustment are convenient, does not produce the damage to road, maintenance is easy, does not affect traffic condition of the road. Owing to the background above, this thesis aims to traffic violation video detection algorithm researching, and focus on two situations of violation: vehicle red light runners detection and vehicle converse.This thesis did some research and improvement in motion vehicle detection and tracking algorithm based on existing research and proposed the algorithm of vehicle violation video detection based on analysis of vehicle motion trajectory . First, the paper used background subtraction to detect the motion vehicle. Background subtraction is based on background modeling and updating, in the thesis, we improved the average filter background modeling algorithm and proposed the method of selective background updating. Then, used Otsu algorithm to Binarize the subtraction image and proposed the noise removing method of binary image in post-processing. Because motion is usually along with the occurrence of shadow, which impacts the correctness and accuracy of motion detection, to detect and remove shadow, we used texture invariance between motion objects and background. Due to complexity of traffic, motion objects not only vehicles but also other kinds of objects, for example, people, bicycle and auto bike. There are big differences between vehicle and other kinds of objects in shape feature, the paper used these differences to classify vehicles from other kinds of objects. Last, we tracked motion vehicles based on region feature matching in different frames and proposed a method of trajectory fitting based on cubic splines interpolation to lay a solid foundation to the method of vehicle violation based on trajectory analysis which proposed in the paper.