The Research On Moving Object Detection And Tracking Algorithms For Intelligent Visual Surveillance

The technology of intelligent visual surveillance is an important application in computer vision area and is applied widely in governments, corporations and families. Then the technology of moving object detection and tracking is the core of intelligent visual surveillance.Over the past 10 years, the research on moving object detection and tracking method for intelligent visual surveillance has attracted lots of researcher, which has been a study hot spot. Now many methods of moving object detection and tracking have been founded, but there are still lots of difficulties, such as background change, object occlusion, shadow, which impact the result of detection and tracking. Therefore it is difficult to design a robust algorithm of detection and tracking.This paper firstly introduces the history and actuality of intelligent visual surveillance; then presents the foundation of image processing and the key technique of the system;meanwhile,the technology of image processing in intelligent visual surveillance has been researched, including image pre-process, object expression and math morphology.Aiming at the application of intelligent visual surveillance system,this paper mainly concentrates on the research on the technique of moving object detection and tracking, then some effective algorithms are proposed.In terms of object detection, some common approaches in the field of object detecting are firstly introduced, including frame subtraction, background subtraction and optical flow. Then the algorithms of background subtraction, direct computing, imaging modeling, combination, are concluded.To solve the problem of background updating, an algorithm based on mixture Gaussian model and frame subtraction is proposed.The model classifies the pixels in each frame into background area,uncovered background area and moving objection area.According to the principle that different pixel area has different update rate, the background model is updated quickly and exactly.About object tracking, some common approaches are firstly introduced:region-based tracking, feature-based tracking, active contour-based tracking and model-based tracking. Then the algorithm based on mean shift is emphasized in this paper,and the element of mean shift is expounded.The paper also introduces the application of mean shift in object tracking. To solve the problem that the traditional mean shift may result in part optimization, a modified algorithm is presented, which is a combination of tracking algorithm and mean shift based on the ideology of detection before tracking. Finally, simulation experiments are given. The results show that the presented algorithm is effective.