Research On Several Problems Of Robot Visual Servoing System

It is an important developing direction that the robot is accompanied by vision so that it can interact intelligently with its working environment. The system of visual servoing can be divided into five subsystems: image preprocessing, feature detection, camera calibration, designing of visual servoing controller and robotic controller. To improve the accuracy and range of application, camera calibration, feature detection and designing of visual servoing controller are deeply researched.In the Second chapter, coordinate transformation,camera projection model and robot control, which are the basic knowledge about visual servoing are described. By visual-feedback signal which is defined in three-dimensional space coordinates or directly in terms of image features, the control structure of visual servoing is classified as image-based and position-based. By analyzing the two control structures, their advantages and disadvantages are presented.In the third chapter, camera calibration method is studied. Because accuracy of the calibration parameters affect directly on the location performance of visual servoing, the nonlinear calibration method which has better accuracy is put forward. It uses genetic algorithm for optimizing intrinsic parameters and gradient descent method for optimizing extrinsic parameters. Compared with classical techniques, the proposed technique is easier, more accuracy and robust, Furthermore, it needn't estimate initial parameters.In the Fourth chapter, feature detection is discussed. Spatial moments not only can accurately describe edge detection and detect subpixel edge feature but alse have rotation invariance. Therefore, by analyzing corner model and spatial moments functions, spatial-moments-generating polynomials are used as one decision rule of corner detection measure. To corner feature, both gradient magnitude change and direction change of gradient vector are discontinuity in a local image neighborhood, Hence gradient-direction variation and gradient magnitude are used as another decision rule of the corner detection to improve antinoise. The two decision rules can detect the corner but can't do the vertex of the corner. Consequently, non-max suppression technique is put forward for...