Design And Implementation Of Video-Based Human Action Recognition Software

As a vital part of computer vision researches, human behavior recognition attracts a growing number of researchers and companies. The rapid development in applications of human behavior recognition, such as:video surveillance, people computer interation, video retrieval and so on, makes the research of human behavior recognition extremely commercial and academic valuable.Human behavior recognition in video has been a difficulty in computer vision researches for a long time. Currently, most of the human behavior recognition are focusing in a single simple act of moving targets under simple scenario. For those complex behaviors under complex real-world scenarios, it is difficult to get a high recognition accuracy now.Human behavior recognition can be divided into three steps:feature of video extraction, training characterization of video and classifications of the characteristics of the video.By referring a large number of papers published in technical journals and conference abroad, this paper did a whole analysis and summary in the development of related area over the world.For the advantages of the feature extraction algorithm and limitations in this research filed, this paper casts outs a new feature which combines multi-threshold pixel switching ration map, edge orientation histograms, motion orientation histograms and local cuboid descriptors, to get a new feature which contains both the local behavior feature and the global behavior feature. In this case, the recognition accuracy is greatly improved compared with the recognition with single feature.What’s more, by using the machine learning method, such as the online dictionary learning and locality-constrained linear coding, we get an improved video feature. And then we combine both sparse linear classifier and linear support vector by scoring mechanism for classification of the video feature representations, and significantly improve the recognition results.Our method is simulated by MATLAB. The recognition results under the method described in this paper are tested under a large number of experiments on videos from the Weizmann video database. K.TH video database, UT-Tower video database, YouTube video database,which guaranteeing a high correctrate when compared with the results from related papers in this research filed.