Research On Moving Object Detection And Tracking In Dynamic Scene

Moving target detection and tracking technology is a core topic in the research field of computer vision. After assiduously studied by domestic and foreign researchers for recent years, a lot of significant theoretical research results and actual application systems have been achieved. However, the complexity of the real environment gives rise to the difficulty of the moving objects detection and tracking accurately. So the research of moving object detection and tracking is still facing many challenging issues. The further research is needed. This dissertation makes a new attempt based on the existing classical algorithms. The main works of this dissertation are presented as following:(1) This dissertation proposes a new model named Block Model for improving Vi Be algorithm which is one of the classical background modeling methods, aiming to the Vi Be algorithm easily brings the problem of Ghost regions when using frame contains moving targets to build model and this regions can’t eliminate as soon as possible. Many neighborhood pixels constitute a block. The improving algorithm uses the discrete cosine transform in the block and the block model was established by the direct component of the discrete cosine transform of every block. Using the block model detects moving targets in the next frame. The experimental results show that the proposed method has good performance in removing Ghost region and high dynamic background.(2) In the taking scenes of moving camera, this dissertation presents a tracking algorithm of non calibration moving target based on spatio – temporal context learning. The presented algorithm randomly selects a number of pixels in the edge of the image to using estimating data about camera shift. It can achieve a forecast image about next frame through these data. The moving objects will detect by using the forecast image compared with the real next frame image. The experimental results show that the method can keep track of target, and has a good tracking real-time.