The Research Of Intelligence Alarm System Based On MPEG4 And Support Vectors Machines

Recently the surveillance and superintend control has received considerable attention ,which is an important complementary and development of Support Vectors Machines theory and neural networks's appliance research. The intelligence alarm system design model is raised based on MPEG4 standard compress method and decision system consisted by Support Vectors Regression and Radial Based Function neural networks in this paper. It can produce high compress ratio and decrease expenses of net transmission because of that the MPEG4 standard uses the codify style based on content and object. At the same time it can locate or abstract the peculiarity object from background availability, So this make it more intelligence in surveying express site. Many corporations and research organizations have developed the CODEC based on MPEG4 standard. But there are many products which assert using the style of MPEG4 only apply it to transit the other format of video picture to MPEG4 format in the video stores phase. It can be said that is effectual to take advantage of its feature to adopt the MPEG4 CODEC in the phase of getting video picture.　Support Vector Machines (SVMs) algorithms based on the foundations of statistical learning theory shows excellent function from basic classification tasks to regression, density estimation, novelty detection, etc. The paper expounds the theoretical relation between the Support Vector Machines classfication and regression which offer a way to classification in regression. The paper accomplishes identification pictures which have small change through the comparison regression function digit of the each piece picture's constant moment invariants based on analysing the support vector regression theory. The Radial Basis Function (RBF) neural network is good at closing function and is a well particular approximator. The paper gives the RBF neural networks training with video image feature which the operator regard it as a safety image or danger image, then use it to finish the second times image identification which is suspected as other in the first time. It proves that the design is operative in alarm system to discover changed image through matlab emulation experimentation.