Research Of The Methods On Distributed Video Key Frame Extraction Based On Hadoop Architecture

The video surveillance technology has been widely used in various fields, which leads to an exponential increase in the amount of video data. People have been tired of the traditional way of browsing. Not only does it consume a lot of time and energy, but also the target information will be missed easily. With the appearance of the technology of key frame, this situation has been greatly improved. Key frames will not be affected by time or synchronization of audio and video, and can provide a variety of ways for browsing and navigation. But in the actual application process of extraction of key frames, the biggest difficulty is that the extraction speed of key frames is too slow:on the one hand, the algorithm is too complex; on the other hand, a standalone mode is used.In this thesis, after full investigation of the algorithm of key frame, a new method of key frame is put forward. We solved the core issues in the process of transferring the algorithm to the Hadoop cloud platform, such as the integrity of the frame. Then the algorithm has been successfully transformed into a distributed extraction model, and the extraction speed of key frames is greatly improved.First of all, this thesis points out the importance of key frame technology for video retrieval and other practical engineering applications, and introduces the development of cloud computing technology as well as its application in the field of multimedia in recent years. At the same time, the core technology and theory of Hadoop are fully studied and researched.Secondly, this thesis analyzes the existing problems in current methods of key frame extraction, and a new adaptive method of video key frame extraction is proposed. The method can adaptively determine the number of key frames, decrease computational work and make the processed results of the content gradient video better.Thirdly, this thesis analyzes the problem of extracting video key frames on Hadoop cloud platform, such as integrity of frame, MapReduce of processing logic and other issues, and the effective solution is given.At the end of this thesis, on the basis of the above theoretical support and the full investigation of the related technologies of cloud application development, the Hadoop distributed cluster environment is built, and the distributed video key frame extraction system based on Hadoop cloud platform is realized. Experimental results show that the algorithm in this thesis can greatly improve the key frame extraction speed. Compared with the standalone mode, it’s more suitable for processing video big data.