Dynamic Vision Analysis Based On Deterministic Learning

In recent years, researches in the aspect of computer vision have got rapid development,among which the analysis of movement for dynamic scene has gained much importance anddevelopment. The research analysis of dynamic scenes has now take the place of static one, andfurther up the special use of visual system will soon be replaced by the research of commonlyusing ones, it's the orientation of computer vision development in the future.The recovery of three dimensional structure and motion from time vary images with theaid of CCD camera(s) is usually performed using a nonlinear dynamic system, often referredto as a perspective dynamic system, The major tasks are formulated as the problem of stateestimation and parameter estimation though using of dynamical system theory. One of thetypical method is the using of observer to estimate the dynamical system states or theparameter, numerous successful research has been reported in this aspect. The persistency ofexciting condition can ensure that the perspective system with constant parameter or not isobservable, thus the observing error system will be stable and the errors will convergeexponentially. The satisfying of persistency of exciting condition make the using of the recentresults on deterministic learning theory with the observing of perspective system becomepossible. Based on the recent results on deterministic learning theory, when the system statesare periodic or recurrent, RBF neural networks can satisfy the partial PE condition along thestates, the system dynamics will be learned by RBF neural networks and saved in a way ofconstant RBF neural networks, and the learning error converges exponentially to a smallneighborhood of zero. Take the constant RBF neural networks achieved as training pattern toform the bank of system dynamical patterns, and before that a similarity definition is given.When meeting new system dynamics which are considered as test patterns, it can be used toachieve rapid recognition of system dynamical patterns between the test and trainingdynamical patterns.Human visual system has amazing ability to analyze and understand knowledge. Thissuperb visual analysis and understanding ability has much relative to the large number ofknowledge accumulation in the brain. How to select, acquire, update visual knowledge, and how to use those knowledge to analyze the scene still far from thorough research. In thisarticle, to learn the knowledge by deterministic learning though observation has formulate adeterministic form for the learning and presenting, recognizing of visual knowledge. Thusby deterministic learning the under system dynamic can be learning and presenting,recognizing in a effective way, the learned knowledge can be used repeatedly, withoutlearning it for an another time. Though it has much different to human learning, it affords avery general method different from special in computer vision.