Global Motion Estimation Algorithm Research Based On H.264Standard

The digital video technology is widely used in the communication and broadcastfields with the development of information technology. The video information isintuitive, efficient and extensive, but the information content is too huge. Theproblem of video coding must be solved to guarantee the quality of compressedvideo for application.In March2003, ITU-T/ISO Project Group published the specification ofH.264/AVC which is used for video compression. It has more excellent performancethan the previous specifications. This paper analysis the coding strategies and thetechnical features of H.264/AVC, especially the global motion estimation algorithm.The improvements of this algorithm are also provided.This paper firstly gives a simple introduction of H.264/AVC video codingspecification and then analysis several key technologies. These technologies caneffectively improve the performance of the encoder, but more computationalcomplexity is involved. Thus, how to reduce the computational complexity andimprove the encoding speed becomes the key point to realize real-time encoder.Global motion estimation needs a specific camera motion model. The parameters ofthis model can get from the background motion information of the video sequence.Global motion estimation is a method which is widely used in computer visual andvideo processing field. Global motion estimation has some shortcomings: The huge computational operation makes it is difficult to satisfy the real-time requirement. Thenumber of noise point can affect the accuracy of estimates. In order to reduce thecomputational complexity of global motion estimation，we can choose reliablefeature points to reduce the computing points, this can speed the operation. Thegradient which is a main feature in global motion estimation, it is affected by noisepoint. If the gradient of the point is small, it has more affects provided by noise point.The smaller gradient will become unreliable. If these points are put in the operation,the amount of operation is huge and the result is not accurate. Those gradient biggerpixels can improve the calculate speed, and has no affect on robustness of thealgorithm. In this paper a global motion estimation method based on the pixelgradient strategy was proposed. With regard to global motion models, we adoptsix-parameter affine model because of its reasonable tradeoff between complexityand accuracy. In order to accelerate the algorithm, two-level pyramid is utilized inthe calculation. It uses those gradient bigger pixels in the calculation of everypyramid level, and combines Gauss-Newton algorithm. The extensive experimentsshow that the proposed method is efficient and robust in terms of both computationalcomplexity and accuracy.In the final part of this paper, some knowledge of DSP and CCS software isintroduced; the realization of H.264encoder on DSP is also attached.The research of this paper has application value which can be used incommunication system such as video meeting, video phone and intelligent videosurveillance.