The Research On Tracking Algorithms Of Moving Targets In Video Images

The moving target tracking in video image is an important research subject in the development of digital video technology. It comes down to many areas such as Video Image Processing, Medical Image Analysis, Biology, Automatic Control, Computer Image Processing, Artificial Intelligence and so on. It is of great theoretical and practical significance.In-depth research into moving target detection and tracking methods was carried out in this thesis, and the improvement on traditional algorithms was listed below.Firstly, the methods of Interframe Difference and Background Subtraction based on Hierarchical Model Analysis are introduced as major methods in this thesis. The scheme of combining the two methods was proposed to detect moving targets. This scenario could effectively avoid the problem that the overlapped part of target in adjacent frame was beyond detection in the former method, as well as the problem that it was vulnerable to external environment in the latter case.Secondly, great emphasis is laid on a widely used algorithm named Meanshift at present, which tended to have the tracking region shifted during the process and might end in failure as the error accumulated. However, the combination of Background Subtraction method based on Hierarchical Model Analysis and traditional Meanshift method was suggested to fight against those shortcomings. Also, the improved algorithm is with good robustness.Thirdly, attention was paid to Camshift method, a traditional algorithm that required manual selection of the target region, and might track unsuccessfully when the color of the target was similar to that of the background. Luckily, the KIM method could be combined with Camshift, as was done in this thesis, to cope with the problem. This method could determine the moving target region with effect, especially for those partially masked targets. When unobstructed, the outline of those moving targets could be extracted relatively in its entirety, thereby the accurate tracking would be realized.Finally, the research on tracking method of Particle Filtering is illustrated. Allowing for the situation that tracking might become invalid when using single characteristic, the color information and the shape information were integrated to describe the target, which was applied to Particle Filtering to overcome its flaws. This method had excellent adaptiveness for part-time masked targets.