Three Dimensional CORF Patio-temporal Model Of Simulating Simple Cells In Visual Cortex And Its Application

Understanding the mechanism of the creatures' action detection is a significant field vary from computer science to brain science to psychology and other cross science disciplines.Action detection is the most challenge problem in computer vision,which is gorgeously applied at the automatic monitoring,human-computer interaction and the like.Human's visual system is the most efficient system for action recognition.Counterfeiting the creatures' visual system to build software model is the emphasis of bionic field for decades,and there product lots of models with gorgeous reference value.But instead of thinking about the pre-processing of the front-end part of visual system,most of them separate the whole construction of the visual system,and only consider the visual cortex and the farther parts.In this paper,we mainly consider the pretreatment of the sub-cortical cell,which means LGN cell.Based on the formation principle of simple cells in real visual cortical,considering the mapping from LGN to simple cells,we realize the model of simple cell,and farther,we have corresponding analysis and application.We've achieved some research productions.Dorsal stream in the visual system is an important mechanism for movement detection and action recognition.We've built a model,which called 3D CORF model,for action recognition used to detect simple movement.It can explain and predict the neuro-physiological behave in the dorsal stream.3D CORF model is a V1 simple cell model relying on the dorsal LGN cells' input.The model includes a spatio-temporal feature detector which can increase complex: at first,the input image sequence is dealt with a series of LGNd cells,then the synthetic sub-units of the whole responses map to be V1 simple cells with special direction,and at last the V1 cells detect the stimulus' direction and speed.The model owns great detect effects,better than the current-fashion 3D Gabor,which is the represent for non-linear models with half-wave rectification.Then,in order to apply the 3D CORF model to human action recognition,we build the action recognition system simulating the mechanism of double pathways in visual system.And in it,the V1 simple cells in ventral pathway employ the 2D CORF model,and the V1 simple cells in dorsal pathway employ the 3D CORF model.SVM is used for classifier.We analyze the received data obtaining from the experiments on the public KTH database,and get two conclusions that one is the system in this paper owns high-efficiency subject detection performance,and the other is double-pathway model can work better than single-pathway model.Both the 3D CORF model and the action recognition system of simulating visual system theory mechanism can greatly explain anatomico-physiologically the processing visual information mechanism in the visual system.They can also preferably help the cross disciplines such as computer vision understand the physiological functions,and accelerate the computer vision field develop forward fast.