Study On Theory And Application Of Human's Eyes And Neck Vision Simulation System

Machine vision is a newly developed, composite and multi-disciplinary subject. It studies on the cognition science of vision information from the level of information processing. Machine vision system is a synthesis of hardware and software, which can perform certain vision functions. After researchers' several decades' effort, machine vision develops rapidly, and it goes into practical application from laboratory. But to be objectively speaking, machine vision is still in immature stage. Either theory or practical application is deficient. This thesis concentrates on the human's eye and neck vision simulation system. It constructs the framework of the simulation system, studies on the basic theory, technology and application. The primary work includes system model, monocular motion analysis under complex background, objects active tracking, versatile camera calibration technology, specific face reconstruction based stereovision and DSP technology based vision processing.In this thesis the human's eye and neck vision simulation system, machine vision theory system and development are introduced. The state-of-the-art of the motion vision analysis and stereovision system is summarized. The existing problem and development trend, which are related with machine vision, are presented.The math models, which include kinematics math model of the human's eye and neck system and math model of the camera, are established. With analysis of the math model of camera in detail, a new camera calibration method based on line rectification is proposed. A linear algorithm is provided to obtain the solution. Compared with non-linear methods, it does not require a good initial guess to guarantee convergence and it does not have the problem of converging to a local minimum. The accuracy is the same as that of non-linear methods.Motion vision analysis is studied in detail. A motion objects detection algorithm under complex background is proposed. Objects are detected through scene Gaussian modeling. A series of methods are proposed to weaken the influence, which is brought by the shadow of the objects and lighting condition variations.A human body recognition method based Support Vector Machine is proposed to distinguish the different possible objects. Feature abstracts based on head and shoulder model and center radiating vector representation are used in different application. Human objects recognition classifier is obtained through samples collecting, feature abstract and SVM training. Experiments show that SVM suits for objects classifying under the condition of limited training samples.It studies on motion objects position, principle of tracking and implement method. The math model of the human's eye and neck system is established. A background match combined with inter-frame differential method is used to solve the difficulty of objects detection under the situation of both camera and objects moving. A Kalman predictor is used to estimate the object's position in order to make object tracking more steadily. Monocular tracking and binocular tracking control methods are also studied.Theory analysis and experiment are performed to human's specific face recon struction based binocular vision. The math model of binocular vision length meas urement is set up. The principle of the face reconstruction based on binocular vision i: introduced. A new energy minimization equation is proposed in stereo matching. Th< disparity is obtained by using correlation calculation with pyramid structure and active contour model. The specific face is reconstructed successfully.A universal vision-processing platform based on DSP technology is developed The human's eye and neck vision processing task can be implemented with this platform. It is a new way for vision processing without personal computers. The systen design, the main processor and peripherals selection, modules and circuit design anc some experiments are introduced in detail.At last, two implementation examples using some achievements are introduced One is the object auto tracking integrated intelligent dome camera, the other is thu status monitoring and fault diagnosis pneumatic system based on vison technology.