Research On Single Object Tracking Based On Video

As the research hotspot of computer vision and digital image processing, object tracking has already obtained many outstanding achievements. However, large changing scales of objects, variation of movement and image blur are the challenging problems of object tracking. One of the key problems that affect the performance of object tracking is extraction and representing method of features. Nowadays, most tracking algorithms are based on textual feature or color feature. In this thesis, superpixels--the hottest feature extraction algorithm in the domain of computer vision currently is to be discussed. And object tracking method based on the feature of superpixel is to be improved and optimized.First, symmetrical SLIC superpixel feature extraction algorithm as well as symmetrical and random SLIC superpixel feature extraction algorithm is to be put forward. Those two algorithms improve the method of superpixel in choosing center of clustering, reduce redundant cluster center, increase use ratio of center of clustering and extraction rate of superpixel. Symmetrical and random SLIC super-pixel abstraction algorithm can improve boundary adhesive rate by adjusting the size of search window according to requirement. A large number of contrast experiments prove that advanced SLIC superpixel can increase algorithmic ratio effectively, reduce memory consumption and improve adhesive rate of boundary.Second, incorporation of flexible searching rectangle and features of advanced SLIC superpixel can improve tracking effect during the appearance of blur images. The flexible searching rectangle is used for extracting features by bringing scale information into objects’ observation model. By integrating features of advanced SLIC superpixel, tracking effect during the appearance of blur images is to be improved. A large number of contrast experiments testify that: ①advanced super-pixel tracking algorithm can resist the change of size and large-scale shape changes. ②by bring in features of advanced SLIC super-pixel, algorithm can tracking blue images effectively.Third, it is shown that our algorithm can efficiently deal with variation of motion, scales and image blur, etc. Our method can efficiently distinguish between the object and the background by combining the improved SLIC superpixel algorithm and deal with the variation of scales and image blur.