Chameleon Motivated Approach Based On The Platform Of Binocular Active Visual System

Computer vision is a challenging and important research field in both engineeringand science field. The platform system based on the binocular active visual monitoringwith the baseline adjustable on circular track,is a new type of visual model. With thelong process of elimination and evolution in the nature environment, chameleon hasformed a unique visual behavior mechanism. In this paper, the key point research isimitating the visual behavior when chameleons preying on the new vision model, andcorresponding to the three phases in the chameleons' preying, the vision technologybased on the binocular active visual monitoring platform is discussed. Base on this,some researches are discussed in this paper:First of all, some details of binocular active visual monitoring platform used inthis paper are introduced, such as, the coordinate system and the calibration system ofthe platform. And then, chameleon visual model based on the binocular active visualmonitoring platform is discussed. The visual behavior in the chameleons' preying aredivided into three phases, as follows, the phase of scanning the environment using twoeyes independently, the phase of focusing on the prey using one eye, and the phase ofpredicting the motion of the prey using two eyes.For another, a "negative correlation" scanning strategy built on the binocularactive visual monitoring platform is established. Then, under this scanning pattern, thetwo branch chains detect moving object in the case of dynamic background within therange of their self visual area independently. The algorithms of Harris and SIFT (ScaleInvariant Feature Transform) are combined together to detect and extract the object;Thirdly, a method based on visual view which is inserted in sequence image pairis proposed to determine the pose parameters and the motion parameters of the targetobject. So, the estimation of3D object's posture based on monocular camera and thedynamic environment is realized.The last, under the binocular model,the Extended Kalman Filter Algorithm isintroduced into predictive estimation of target object's moving in the scene. The above algorithms are implemented in the Matlab workbench. And theexperimental results shows that the algorithms proposed in this paper achieve thedesired effect.