The Realization Of Gait Recognition System Based On Deterministic Learning

Gait recognition is a newly developing technology of biometric identification. Itcharacterizes in non-contact, non-invasive and elusive and it overcomes the defects ofrequiring high-definition picture and close contact in traditional biometric identification suchas iris identification and fingerprint identification etc. Gait recognition has good effect underlow resolution circumstances, which makes gait recognition can be widely applied to manyfields such as smart monitoring and human behavior analysis. Gait recognition analyses thevideo picture of human walking and the key point is to extract suitable static or dynamic gaitcharacteristic or its combination to describe human walking characteristic and derive gaitlabel together with the sorter to recognize. Gait recognition combines technologies ofdifferent fields such as computer vision, video image processing and pattern recognition etc.Deterministic learning is a new theory which applies to machine learning under dynamiccircumstance in system and control field. It applies the concepts and methods in the fields ofself-adaptive control, RBF neural network, and dynamics system etc. and based on which itpresents a good way in the expression, the similarity definition, the fast recognition andclassification of dynamic pattern and intelligent control via dynamic pattern recognition. Weclassify human gait recognition as a problem in dynamic pattern recognition and apply themethods of dynamic pattern recognition in deterministic learning theory to human gaitrecognition.Though the research to gait recognition is still on theory level so far, many scholarshome and abroad have taken much energe in this field and obtained many achievements. Atpresent, there are few products that gait recognition can be applied to. What’s more, thereexists great difference between research achievements and pratical application. This paperintroduces the theoretical basis of gait recognition system based on deterministic learningtheory, on which this paper highlightly discusses the system’s realization that is the realizationof extrat characteristic data used for gait recognition rapidly, automatically and accuratelyfrom the human motion picture sequence and the realization of gait pattern practice and gaitrecognition based on deterministic learning theory. The preliminary realization work has been completed by combining OpenCV and LabVIEW. Furthermore, this system has beenexperimentized on UCSD gait data base, self-created gait data base, CASIA Dataset A andDataset B and acquired high recognition rate.