Online Video Analysis Technology Research Based On Machine Learning

Video data is important representatives for information storage. However, highsimilarity always exists among the video frames in many cases, such as some kind ofsurveillance videos, and so on. This will lead to information redundancy, and furtherwill cause problems in information storage, query and transfer. In this paper, we presentan algorithm for video semantic enrichment. This method can perceive the videocontent quickly and accurately by extracting the key frames from video sequences. First,we extract the continuous foreground segments from the original video sequences toremove the background frames. Then, we segment the foreground to be a series of subvideo sequences by the shot boundary detection process. The features extracted from thesub video sequences constitute the original dictionary. Finally, we select the key frameswith a dictionary selection process. By removing the useless information with abackground model, our method can deal with videos of arbitrary length by the shotboundary detection. The meaningful key frames can be extracted based on the imagefeatures and dictionary selection method.By using C++programming language and Open CV library, we write a programwith MFC interface. The program can automatically extract key frames from the inputvideo. In addition, it also displays key frames and scores and saves them. In the last partof the paper, compared with the artificial identification method and the baselinealgorithms, the experiment results verify that our method can obtain more accurate keyframes.