The Technology Of Human Movement Recognition

Typically, intelligent video surveillance system is made up of a variety oftechnologies; they are moving target detection, target classification and identification,target tracking, behavior understanding and description. As the foundation of themonitoring system, the mission of target motion detection is extracting the regions ofvariation as accurate as possible. This is regarded as the key and difficult point in themonitoring and computer vision fields at home and abroad. Detection of moving objectsplays an important role in the area of follow-up advanced processing, it performancewill have a direct impact on the timeliness, accuracy and robustness.The abrupt change of light intensity and the varieties of background make barriersto accurate moving object extraction, the background extracting and update becomemore difficult too. GMM which is raised by Stuaffer and Grimsion is one of the classicalgorithms strong the high anti-jamming ability and robustness. Under the disturbanceor strong noise, GMM will have miscalculation. In order to solve the shortages, thispaper puts forward an improved method after further study and analysis.1. Changing the background model matching conditionsMatching conditions between pixels and background model involving pixels, mean,variance and standard deviation according the classic GMM raised by Stuaffer andGrimsion. This lets the variance into a small value and the background a little distortion.Background is more closer to real condition because of the improved matchingcondition just involves the pixel and mean.2. Adding lifetime and reproduce frequency to background modelThis system will add lifetime and reproduce frequency to classic GMM. The twonew parameters can be used to reserve the model which has short lifetime and highreproduce frequency. This new model can accommodate high-frequency noise, such asthe blowing leaves or the flow of water.3. Foreground extraction based on Background subtractionThe prospects extraction in this paper uses the combination of improved GMM andbackground subtraction. The new GMM can analog simulation the background stablyand well. Simultaneously, background subtraction method has a strong adoptioncapacity of scene change, a better inhibition of the shadow of the target. Combining newGMM with background subtraction method can make more precise and robust inforeground extraction.