Research On Multiple Objects Tracking Based On Detection In Video Sequence

In recent years, intelligent video surveillance has become one of the research focuses in computer vision field. It has been widely used in intelligent traffic monitoring, building safety and other fields. Object detection and tracking are not only the key technologies in intelligent video surveillance system, but also the foundations of object recognition, behavior analysis and so on. In this thesis, multiple object detection and tracking technology under single fixed camera are studied. The content is composed of object detection and object tracking.In object detection, firstly, three classic object detection algorithms are analyzed. Combining with the specific surveillance environment, background subtraction algorithm is used to detecting the moving objects based on the background which is built with frame difference and updated continuously. Then, the optimization process is discussed, such as image binarization and morphological filtering.Object tracking is divided into single object and multiple objects. When tracking a single object, background weighted color histogram is used to improve the robustness of MeanShift tracking algorithm. While tracking multiple objects, a new algorithm based on object status is proposed. With the idea of the segmentation of tracking process, objects are divided into five situations: single object, object appearance, object disappearance, object integration and object separating. Then, object matching matrix and multi-feature tracking algorithm are integrated effectively, so that the tracking accuracy is improved and the complexity of algorithm is reduced.The results of experiments show that the algorithm has a better accuracy and real-time performance when monitoring multiple moving objects.