| With the rise of the internet of things and “smart city”,the scalar data obtained by the traditional Wireless Sensor Network(WSN)can no longer meet the application requirements,and gradually transition to multimedia data acquisition such as image and video.Therefore,Video Sensor Networks(VSN)came into being.Video Sensor Networks are widely used in transportation,security and environmental protection.While helping people collect a large amount of video information resources,it also puts tremendous pressure on data storage and network transmission.In order to solve this problem,video compression method is needed.Compared with traditional video application scenarios,the new video application scenario of Video Sensor Networks has different characteristics: it has a large number of video sensors,limited storage capacity,low data processing capability,and limited battery life.Therefore,the traditional high complexity coded video coding standard is difficult to apply to the Video Sensor Network,and thus the high coding complexity is difficult to implement on the video sensors.If the uncompressed raw video data is transmitted to the central server and compressed with traditional high complexity coding,such transmission process will consume a lot of bandwidth and energy,which is inconsistent with the original intention of video compression.Therefore,the new Video Sensor Network application scene requires a video compression coding method with low coding complexity,higher compression efficiency,and better compression performance.As for decoding,high-complexity decoding can send compressed video to the control center for decoding on a central server or PC.In view of this,this thesis adopts an improved distributed video coding based on trellis coded quantization for video coding compression.This thesis combines the characteristics of video sensor networks,video application scenarios and final application purposes,and studies how to compress the video data generated by the video sensor network.The focus of the research is to extract the foreground moving objects first,and to realize the separation of the foreground moving objects and the background.Then,the foreground moving objects adopt the distributed video coding based on grid coding quantization for video compression,and the background has a general high compression ratio intraframe compression.This thesis mainly studies from the following aspects:(1)According to the data features of the Video Sensor Network,the background difference method is used to realize the real-time detection of foreground moving objects.(2)Analyze the common modeling methods and propose an improved hybrid Gaussian background modeling method to construct the reference background of the video.(3)Firstly,the characteristics of background noise are analyzed,and then a detection method based on macroblock classification is adopted to realize the preliminary determination of the foreground moving target region,and the uncertain macroblock is used to accurately detect the background noise model.Finally,use the results of macroblock classification to construct a pure background image and extract moving targets.(4)Based on the original PRISM framework,the original distributed video coding algorithm is optimized and improved,and a distributed video coding algorithm based on trellis coding quantization is proposed.The foreground moving object video data is then compressed using this improved distributed video compression encoding algorithm.Through experimental analysis and comparison,the performance of this designed VSN-oriented distributed video coding method is verified,and the conclusion is: The video compression method designed in this thesis has a higher compression ratio than the original MPEG-4 encoding method,and the amount of data after compression is less.In addition,in the compressed video frame image,the foreground moving object image reconstructed image has good quality,high fidelity and no distortion;the background image has a certain distortion,resulting in a block phenomenon.For cameras and surveillance,the distortion of the reference background can be ignored as long as it does not affect the visual effects and subsequent research of the foreground moving object area. |
PDF Full Text Request