Research On Moving Object Abstraction And Track Algorithm For H.264 Etc

Nowadays, the demand that people want to search and index video's content becomes more and more urgent, especially for the motion object's abstraction and track. Because most of the video data is archived in compressed domain, people tend to do abstraction in compressed domain directly, which can save plenty of time for decoding the video.By now, many scholars have done some research work for this topic, and proposed many effective algorithms, but most of those algorithms are aimed at the First Generation Video Compression Standards (MPEG-1, MPEG-2 etc.). The Second Generation Video Compression Standards (H.264, AVS etc.) have have introduced many novel and advanced algorithms which achieve plenty of compression efficiency, but on the other hand make the traditional abstraction algorithms useless, so the abstraction algorithm for the 2nd video standards must be researched. This paper funded by Hubei Science Foundation (No. CGZ0223 ) researchs the motion objection abstraction and track algorithm for the 2nd Video Compression Standard.Aimed at the 2nd Generation Video Compression Standards, this paper proposes a novel and effective motion object abstraction and track algorithm. This algorithm includes a new vector median filter algorithm: Separate Vector Median Filter (SVM) which can gain good filtered result with much lower computation complexity. After that, an abstraction algorithm using mv gobbet's temporal correlation is executed which can get an exact result with high robustness. At last, a method manipulating intra-coded block was introduced, which can judge these blocks to appropriate motion objects or background. The experiments show that the proposed algorithm can get a precise result.Besides, based on the motion object abstraction, a motion object track algorithm using motion reference is proposed. In the course of object track, motion objects are often lost, so this paper submits a method to manage this situation. This method treats the object's motion in a short period as uniformly accelerated motion, and predicts the lost motion object's current position using its positions and motion vectors in previous two frames. As the experiments tell us, this track method achieves a good result.