The Research On Moving Object Detection And Tracking Based On Traffic Surveillance Video

The moving object detection and tracking based on video have a broad application prospect in many fields such as security surveillance, intelligent transportation, navigation and so on, thus they attract great attention from the scholars in the field of computer vision. Take the Intelligent Transportation Field for example, the traffic surveillance video provides the real-time traffic information on road in wide-area scenes. Through detecting, tracking and identifying the moving objects in the surveillance scene, we can obtain the traffic information, such as traffic flow, speed, road capacity, which can offer some scientific reference for the traffic controlling, induction and signal timing decisions and consequently make the urban traffic network operate efficiently and safely.Taken the single junction traffic surveillance video as the research object, this thesis explores the issue of the moving object detection and tracking of the traffic surveillance video in a stationary monitoring camera. The main content and results are summarized as follows:(1) After analyzing the background modeling issue of the video taken from a stationary camera, we introduce a new model based on Effect Components Description, and use the Mean shift algorithm to locate the modes of the underlying distribution pixels in the image sequences to generate an ideal background. This Model is robust to the interfering factors, such as the noise, the vibration of the camera, and can extract a very clear background without blur effect from the image sequences with cluttered moving objects and the result is better than the commonly used Mean Filter Model and Gaussian Mixture Model.(2) Analyze the potential defect of the IIR filter background updating algorithm in practice and propose an approach based on two frames which can avoid the defect of IIR filter that cannot reflect the real monitoring scenario. Meantime, considering the read-time and robust demand of the algorithm in practical application, we illustrate a background updating strategy using various algorithms based on time-period segmentation combine according to the advantages and disadvantages of the algorithms presented.(3) Due to the defect of the issue that the image segmentation algorithm based on Maximum Entropy has a large amount of calculation and is time-consuming, we proposes a new way based on linear scale compression of the image, and directly segments the images whose gray scale was compressed. The results of the experiments show that the effects of the object segmentation are not affected. This method reduces the amount of computation effectively, and can improve the real- time processing.(4) In the phase of object tracking, an algorithm combine with gray prediction and multi-feature matching was utilized. The gray prediction model was established through the moving object location data to predict the position of the moving object in the next frame, which can reduce the search area for target matching. Considering the defect that sometimes the gray prediction is unstable and has deviation, we introduce the Markov model to revise the prediction deviation and improve prediction accuracy. Then, the real-time tracking of the moving object in the traffic surveillance scene can be implemented by feature matching operation which extracts from the corner and part of the global features of the moving object.This thesis studies the detection and tracking issue of the traffic surveillance video taken from a stationary camera. Specifically, we discuss the algorithms of background modeling, background updating, object segmentation, prediction-track, and finally propose some effective solutions. On this basis, adding to the help of the OpenCV's framework, the study proposes a system for moving object detection and tracking which can detect and track the vehicles in the video.