Research On Object Extraction And Tracking Based On MPEG-4 Video Objects

An important source of information that humans acquire is vision. Scientific study indicates that, of all the information that human brains gains, three quarters of these are image stream. In the mean time, along with the rapid development of technology, information of multi-media becomes more and more which far more from the ability of humans that can handle. Consequently, it is an urge that multi-media information should be operate automatically by computer, such as auto-sorting depending on different attributes of videos, manipulating objects in video stream by response to human command and detecting and tracking in real time the object that people interested and etc. Thus, technologies that are able to acquire and process relevant images, video streams to find information that we needed is difficult and is the core of these studies.Facing with these coming needs, traditional video compress method, although they could gain much higher compress rate and applied in much applications, such as MPEG-1/2 and H.26x, are based on frame technology, and do not divide and extract objects from sceneries. Thus, they could not satisfy needs. Therefore, the MPEG-4, which using a different method and coding standard different from that of based on frame compression, is based on coding method that video object plan which contain many video objects and which is the basis of video object tracking.In order to study the objects in a video sequence, it is therefore first need to extract what we find interested and then find a way to describe these objects and the third is to tracking them using some algorithms. However, it is hard to divide and extract video objects. Meanwhile, different tracking method is used when method that describes these objects varies.Thus, after study in different main stream methods of divide, extract, describe and tracking method, a new method that combines interests point and geometry points to represent an video object is presented in this paper, which render more accurate in object description and more robustness in occlude recovery. What's more, the number of interest point, which may be a great amount and many irrelevant ones, is huge. For these reasons, a new approach is represented in this paper, which may decrease dramatically the irrelevant interest point and reduces the redundant ones in video object itself. Method that tracking object using this description is given in this paper at the same way. Due to the process used above, the cost of calculation Mahalanobis distance is reduced.Finally, the validity of these methods that divide and tracking video objects is proved by experiment. The result indicated that, the method could automatically find and tracking video objects with many irrelevant interest points been eliminated. And the object tracking process using the algorithm also achieves better results.