Research On Moving Object Detection And Tracking In Dynamic Scenes

Moving object detection and tracking is an important and fundamental problem incomputer vision. To improve the object detection speed, integrity and tracking robustnessunder dynamic scenes, this paper researches some moving object detection and trackingalgorithms, and then proposes two improved methods. The main works of this paper are asfollows:Firstly, in order to increase the speed and integrity of moving object detection in dynamicscenes, an algorithm combining CenSurE and spatial-temporal information is proposed.Experiment results show this proposed algorithm could achieve the speed of15frames/s, andit also obtain integrity of moving objects at the same time. The proposed algorithm basicallymeets the demands of moving object detection in dynamic scenes in terms of speed, noiseresistance, light adaptability, object integrity and so on.Secondly, in order to track object continuously and accurately even in the scene withsimilar color and illumination changing, a new local saliency texture descriptor named LocalTernary Number (LTN) is defined, and a new Mean Shift tracking method combining LTNwith hue information is proposed. The experiment results demonstrate that the proposedmethod improved the performance of traditional color-based Mean Shift tracker and it cantrack object consistently and accurately in the above scene.