Research On Video Segmentation And Motion Recognition Algorithm Based On Global Motion

With the improvement of living standards, people are having more and more high qualityrequirements of life, the pursuit of green, healthy life. However, the reality of the situation isoften a difference for many people. In modern society, more and more diabetes, hypertension,obesity and other wealth diseases appear. However, through the analysis of someexperimental preliminary findings, the cause of these diseases, mainly due to excessiveenergy intake and too little exercise to burn energy, less-output and multiple-input lead to theresults of the various diseases. The research of wealth disease prevention has been going on athome and abroad, and wearable video provide a lot of support and might to the conduct ofsuch research. In this paper, the experimenters shot the video information through thewearable video, and the videos are regarded as a record of the experimenter life phenomenon.By analyzing the video, it can segment experimenter day activities, and then make a motionrecognition.Expand the analysis and discussion in this paper, mainly from the following aspects:(A) Extraction of video key frameBy extracting the key frame of experimenter life videos, it can make a preparation for thelatter part of the video analysis and processing. Wear video features a button camera, worn onthe chest of the experimenter, shooting video with a "no lens" characteristics, that is to sayany angle shooting, the subject does not appear in the video. Therefore, this paper proposes asegmentation method based on global motion. It regards each video image in each frame ofvideo as a whole for processing, and no longer a separate deal with the subject.(B) the global motion-based on video segmentationThe traditional video segmentation is mainly divided into two types, one is based on theseparation of the subject and the background, the another is based on the lens. In the test ofthe present paper, the wearable device is worn on the human body, and the body imaged cannot appear in the lens, so there is no separation of the subject and the background. Further,in the acquisition process of the experimental data, once the lens opened, it will not be closedunder normal circumstances, resulting that there is only one lens in a video file. In the videofiles with only one lens, the lens cannot be deemed to be the basis of video segmentation. So,when dividing the surveillance videos shot by a wearable device, it needs to re-select theother split basis. The movement of the wearable capturing device and the body of the humanmonitored is the same, so the motion of the video is just a reaction of the body motion.Motion-based video segmentation can segment the videos in accordance with the state ofmotions, so that the monitored human’s daily movement can be gained and do furtheranalysis.(C) Indirect human motion recognitionSince there is different energy consumption in different state of motion, the paper shouldestimate the motion state of the monitoring video human in order to count the energyconsumption. In view of the electronic equipment is consistent movement with the movementof the people, the video shot of the scene changes can reflect the movement of human. In thispaper, namely using the mapping relationship between the camera movement and the humanmovement, and by analyzing the motion characteristics of the video series, this paper presentsa method to identify the human form of exercise without having to look for a sign in the mainvideo. In this paper, by comparison the pros and cons of the two methods-indirect way ofhuman motion recognition based on optical flow and diamond search block matching methodbased on improved to movement recognition, it can make the motion recognition correctly.After the experimental data, this paper can know the accuracy of the estimates of theabove two methods-video segmentation and motion recognition, and can also show thepractical significance in reality.