Low Delay H.264Video Encoding Key Technology Research For Surveillance Video

Along with development of video technology, surveillance video ismore and more important for public security. Because H.264/AVCstandard has high encoding efficiency, it is widely used in surveillancevideo applications. Low delay encoding is becoming more and moreimportant in surveillance video applications, as a result, low delay H.264video encoding for surveillance video is an important research subject.There are mainly four sources for encoding delay including capturedelay, encoding process delay, frame reorder delay and encoding bufferdelay. Among these four kinds of delay, capture delay depends on thevideo capture systems and frame reorder delay can be0when no Bframes is used for H.264/AVC encoding. The other two kinds of delay,encoding process delay and encoding buffer delay, is the major problemin H.264/AVC low delay encoding technology.H.264/AVC encoding process delay is time latency from videocontent entering encoder to the end of entropy coding. To reduceencoding process delay, slice-level parallel encoding is an importanttechnology and workload balancing is the key technology for slice-level parallel encoding to reduce encoding process delay. In this thesis, wepropose two algorithms for slice-level parallel encoding workloadbalancing. The first algorithm is dynamic slice-resize algorithm forslice-level parallel encoding. This algorithm applies slice-level parallelencoding which can be used for low delay encoding. After applyingdynamic slice-resize algorithm, it balances slice workload and improvesencoding speed. The second algorithm is dynamic slice-resize algorithmwith QP compensation. On the basis of the first algorithm, this algorithmadds QP compensation model to make it more efficient for slice workloadprediction and balancing. Implementation on embedded system for thisalgorithm shows that this algorithm can fits different platforms.H.264/AVC encoding buffer delay is time latency to solve theproblem when bitrate after variable length coding enters fixed ratechannel. To reduce encoding buffer delay in constant bitrate channel withlow bandwidth, it is important to make bitrate stable, however, there isbig difference for bitrate between I frame and P frame which is almostimpossible to make bitrate stable. As a result, dispersal of intra predictionin each frame is a good solution, thus, intra refresh technology is solutionfor big difference for bitrate between frames. In this thesis, we propose anintra refresh algorithm for low delay H.264/AVC video encoding. Thisalgorithm applies forced intra refresh technology and to make sure allMBs can be refreshed and decoded correctly, MB prediction direction is limited. Only the region that has been refreshed can be the reference.Based on this, we use motion adaptive slice-level intra refresh directionselection algorithm to cut down the encoding efficiency reduction causedby limitation of prediction direction.In this thesis, three algorithms are proposed. Dynamic slice-resizealgorithm for slice-level parallel encoding makes workload morebalanced between slices and reduces encoding process delay. Dynamicslice-resize workload balance algorithm with QP compensation makesworkload more balanced and reduces encoding process delay. Motionadaptive slice-level intra refresh direction selection algorithm for lowdelay video encoding offers the ability to start decoding at any time andmore balanced bitrate. This reduces the encoding buffer delay.