Research On Research On Recognition Of Limbs Motion By Machine Vision

Now,video recognition is a new high-tech products what has been more and more people’s attention.As a part of video recognition,the research on human-computer interaction,video surveillance,and motion analysis of video and image is a hot research direction in recent large companies,and it has great development prospects.But now a lot of research of video recognition are used in the computer and mobile phones is used the shooting equipment by people,penetration is very high,but using in mobile phone video recognition research is very small,and mobile aspects Video recognition and computer video recognition is very different what mainly reflected in the mobile phone video than the computer shooting video jitter,noise and more problems.Therefore,this paper discusses these issues and study the video by camera,and I hope to designed to build a mobile phone camera what can capture the video to identify the corresponding action system.In order to build the above-mentioned system,the main research work in this paper as follows.In the aspect of obtaining the foreground image by video,the video is captured by the mobile phone,but the video is easily disturbed by the noise and the background image is not completely still.Through the analysis of the three methods,the corresponding foreground extraction method is calculated.Finally,by comparing the precision of these foreground extraction methods,the Gaussian mixture model is selected as the foreground extraction method.In the aspect of extracting and processing the corresponding target motion features,this paper has the corresponding noise,target motion video length and frame number length problems.The obtained data have the characteristics of "singular point" The extracted foreground pictures are processed by corresponding domain processing,resampling,singular point processing and so on.In this paper,we compare the three methods of random forest,Iterative Decision Tree(GBDT)and Support Vector Machine(SVM),and build an experimental platform to validat e this model.By using this mode what is constitute of Platform for our more than 500 groups of video and 3 action recognition research and analysis.And it is filled with the test sample with a ratio of 3: 7,which means that this part of the sample is related to the training sample,the proportion of 7 means that this part of the sample with the training sample is completely irrelevant.By calculating the satisfaction degree of the user to our recognition effect,we compare the satisfaction of several machine learning methods and get the corresponding optimal algorithm.And through the improvement of the corresponding algorithm,we get the overall framework of our system.Finally,it is concluded that the combination of Gaussian mixture model and GBDT has the best recognition effect,and the recognition rate is more than 95%.In summary,the final decision of the overall system made by this program.