Research Of Motional Objects Detection And Tracking Based On Video Image

Video surveillance device is used on more and more occasions as people's awareness of safety increases, however conventional video surveillance device needed to be observed extendedly by special operator so as to find the possible exceptions which is not only a waste of manpower but also sometimes accompanied by some inevitable man-made errors. Intelligent video surveillance is receiving increasing attention of scholars at home and abroad, and great progress has been made in this field.After reading a plenty of related documentation, in this thesis a research on video image processing in intelligent surveillance system is made. The main contents include building background model, detection and tracking of motion objects. In-depth research is placed on Mixture Gaussian model and background modeling by means of Bayes classification, both of which can offset the effects on background models resulted in indefinite factors involved in background modeling such as light change, shaking branches, position change of background object and so on, especially for outdoor complex situations, and their validity and robustness has been verified through a mass of experiments.Research on detection and tracking algorithm of motion object is made after the appropriate background models are built. The motion object is obtained through background segmentation, then the concrete position and size information of each object is obtained through analysis of connectivity of the motion object. Multiple objects tracking problem can be solved by means of Mean Shift tracking and particle filter algorithms with the color information of the object, and in this thesis an analysis and comparison of the characteristics and differences of the two algorithms through a mass of video image experiments so that the conclusion that good stability and fast operation speed can achieved when Mean Shift method is applied to track the objects with great color difference while particle filter method can implement precise tracking of the motion objects in complex situations can be made.A large number of experiments show both background modeling method and motion object detection and tracking method characterized by better rapidity, stability, and robustness promises an important application value in video surveillance system.