The Research And Implementation Of Video Compression Algorithms For Video Sensor Networks

In recent years, with the development of embedded technology and Internetcommunication, wireless sensor networks are not satisfied with simple scalar datacollection and begin to gradually shift to complex multimedia data obtain. As the latestapplication of wireless sensor networks, video sensor networks’ several characters likeeasy to install, lower cost, can be organized dynamically make it possible to be appliedto each domain of smart video surveillance such as smart transportation, public security,environment surveillance, and emergency aid.At the meantime of video sensor networks obtains lots of video information for us,it also produces vast amount of data which makes it very hard to transport ans store thevideo data. To be fair, when general video compression methods are doing thecompression, they overlook the feature of the surveillance video and the final applygoal, only focus on the data size after the compression. In this paper, I proposed a newobject-based video compression method based on the character of video sensornetworks and the application of surveillance video.First of all, compared with several common moving object detection method，Ichoose background subtraction method which has a better real-time and comprehensiveperformance. Since the quality of background model decides how the backgroundsubtraction work, this paper analysis the some common background modeling method,especially on Gaussian Mixture Model, and it finds out and improve the deficiency oftraditional Gaussian Model.Secondly, in order to improve the speed of motion detection and reduce the effectsof noise on the test results, I improve a detect method based on macroblock types tomake the classification more accurate through hypothesis testing and background noisemodel. Then use the classification results to extract moving target, build purebackground image and separate the moving target and the background.Finally, compress the moving target and background separately. Since I analyzedthe standard objected-based MPEG-4encoding technology and coding framework, apowerful compressor and open-source encoder XviD was selected. According to theaim of video surveillance, we compress the moving target and background differently.Besides, moving target is the main research object, to try to keep the compression distortion, using shape coding, texture coding and motion coding; while background isnot so important, it can make drastic compression. Experiments show that methodproposed in this paper has a better performance over MPEG-4, and decompressed videoimage reconstruction with good visual effects. So this method meet the demand forvideo compression on video sensor networks.