Multi-target Visual Detection And Tracking Research And Development Of Video Surveillance Software Platform

Visual object tracking is an important research topic in computer vision. As a key technology, it is of great value in a large number of applications such as intelligent video surveillance, intelligent traffic monitoring, human-computer interaction and intelligent navigation.In this thesis, we use the motion detection method which includes three steps: background modeling, shadow removal and blob detection to detect moving objects. Three different background modeling methods:running average, mixture of Gaussian and code book are applied to adapt different kinds of video scence. Four different shadow removal methods are introduced to remove the moving shadows in foreground images. Then, we use a special and efficient morphology operation to filter shadow removed foreground image which has a strong ability to remove noise and connect separate parts of one object. After morphological operation, the bounding boxes of blobs in each filtered image are extracted, among which some with little area are filtered.Then, global feature based tracking method and shape based tracking method are proposed. The global feature based tracking method makes use of global features such as color histogram, fuzzy color histogram, histogram of oriented gradients and LBP texture to represent the object appearances. Object appearance matching is used to match objects in frames, and an object moving state model is also introduced. The experiment results show that our global feature based tracking method is very effient. The shape based tracking method is mainly for tracking head contours in bird's view scence. A real-time and robust circle detection method based on randomized Hough transform is introduced to detect the pedestrians' head contours in video frames. A special tracking method is demonstrated to track the circular head contours detected by the circle detection method. The experiment results show that our method has a good detection rate and tracking performance.Finally, an object tracking software framework is introduced to handle the large amount of algorithms and their parameters and their parameters in the whole object tracking method. Evey algorithm is implemented as a separate module and can be configured in the software framework. And modules, parameters and the module execution order can save to fle and load from file.