Image Segmentation And Target Tracking Method Based On Contour

With the continuous development of the video technology, the target tracking which is the basis of video applications has attracted more and more attentions. With the further studies on object tracking, and its use to continue to be spread, people are no longer satisfied with only obtaining the location information of the object in a simple target tracking process, and more information are wanted to know. In some areas of image processing, such as gesture recognition and reconstruction of three-dimensional graphics etc, besides the location information of the target, the shapes of the target are also wanted to obtain, which is prepared for the subsequent operations. Active contour model is just to meet people’s requirement, its essence is partitioning the interested object in the image sequences.The quality of the image segmentation results on the subsequent tracking process has a direct influence, the computational speed of algorithm determine if it applies to real-time video system to track the interested target. In this paper, under the premise to ensure a good segmentation results, how to improve the algorithm speed to meet the real-time system requirements is studied. The mainly research as follows:(1) In this thesis, the principle of traditional snake model is analyzed, the strengths and weaknesses of the model are discussed, on the basis of this, we study on GVF Snake model, which is an improvement method of snake model.(2) The level set method is studied, according to its numerical implementation we analyze why its amount of computation is large and computational efficiency is so low.(3) In this thesis, we focus on study of two geometric active contour model, which are the Li model based on edge information and the C-V model based on regional information. The advantages and disadvantages of the two algorithms are analyzed. And through the experiment, the results of the two methods are contrasted.(4) Studying on the fast level set algorithm, combined with the C-V model to re-design of the speed function, remove the manual process which is used to set the segmentation threshold. Adopt a single list to indicate the contour curve to realize the curve evolution. Under the premise to ensure a good segmentation results, the segmentation velocity is improved. And it is applied to track the contour of interested target, through experimenting on the actual data, the validity of this method is proved.(5) An image segmentation system is achieved on VC++platform, which is used to verify the effect of a variety of segmentation methods studied in this paper.