Research On Video Semantic Understanding Of Online Based On Machine Learning

Video as a popular entertainment way to human is playing an increasinglyimportant role in the life of human. The number of human-generated videos every daybecomes more and more huge.Video in all spheres of society is playing an increasinglyimportant role. So the video analysis and research is very necessary to human, thework has a good role in promoting scientific progress of the whole world. ChineseAcademy of Sciences in Shenyang, vigorously under the auspices of the provincialLiaoning ministerial task "robots online scene perception and understanding", thepaper studied and analyzed video semantic understanding technology.Firstly, for the understanding of the requirements and specificities of videotechnology, to do the research related areas. Principles include machine learning, videoevent model.Secondly,we present an approach to abstract the key frame ofcategory-independence video in this paper. The method based on machine learning ofthis paper will abstract the key frames that stand for the video. In contrast withtraditional key frame abstraction methods, the method mentioned in this paper derivethe key frame not for one type of video. To accomplish this work, at the begin ofexperiment we derive the GIST feature using the GIST descriptor, next make acategory sparse model to judge the importance of each frame, then select the framewith high importance score as the key frame. The frames which are selected will beused to summarize the video.In this paper, we only need one model to accomplish theextraction, which avoid the trouble train a new model for each category.Finally,According MATLAB modeling is easy to control the specificity of theproposed methods, related experiments. Get good experimental results.And use thesteering servo controller and real-time tracking camera for the detection of humanfaces.