Research And Implementation Of Traffic Flow Detection Based On Computer Vision

This thesis focuses on traffic flow detection. In present methods, the video detection method is most studied. Video detection is based on the knowledge of machine vision. Machine vision is developed on the basis of image processing, and widely used in computer science and engineering, signal processing, physics, applied mathematics and statics, neurophysiology and cognitive science research, manufacturing, testing, documentation, analysis, medical diagnosis and military fields, getting broad application prospects. On the basis of previous work, this thesis focuses on the following three major tasks:(1) Studied a background modeling algorithm. According to the characteristics of the target moving objects, this thesis compares the various detection methods based on analysis. The background subtraction methods are used to detect traffic flow. In background subtraction, the most critical aspect is the reconstruction of the true background, the accuracy of the traffic detection is closely related with the quality of the background. On the comparative analysis of the various background modeling method, this thesis propose a new background modeling algorithm to reconstruct the background, lay the foundation of accurately segmenting the moving targets.(2) Studied a shadow removal algorithm. In real scenario, the environment is changing with time. In strong light, the path of vehicle will have shadows, If two vehicles are too close, the shadow may connect them to be one target and may expand the area of vehicles, thus the accuracy reduces. To the characteristics of shadows, this thesis analyze of the differences between the shadows and the prospect of vehicles, and on this basis propose a shadow removal algorithm.(3) Studied a motion tracking algorithm. Vehicle tracking is important in the Intelligent Transportation Systems. Vehicles’ tracking is the foundation of its behavior analyzing. After analyzing the characteristics of the target’s motion, this thesis combined the traditional feature tracking algorithm and motion prediction, getting good result.Each algorithm this paper proposes is implements using the C++ language. Amount of experiments show that the algorithms achieve good results.