Study On Fundamental Theories And Applied Technique Of Computer Vision Inspection

The dissertation is drawn up during the work of the two following projects , "the study of visual measuring technique" which is supported by doctoral fund of State Education Commission and "the research on the fundamental theories of visual testing" which is supported by the National Science and Technique Committee. Aiming to solve some key problems in the computer visual measurement , which includes the signal processing methods of the visual sensors, the algorithms of recovering the "feature point" and "feature line" at sub pixel accuracy, the calibration of the visual system and the positioning of the objects, the author made comprehensive studies on them both theoretically and in practice. After all some important theories and methods of visual measurement and testing are creatively presented, and proved by computer emulation and experiments The main and creative work in the dissertation includes the following aspects.1. The basic construction and the property of the computer visual measurement and testing system are generally analyzed. In addition to introducing the evaluating criterion for the property of the visual system, the author definitely points out that key factors which lead to an satisfactory visual system are the accurate signal processing, the proper methods of calibration for the system and the precise positioning of the measured objects.2. Originating from the principles of projective and perspective transformation, the models of commonly used visual sensors (typically, the line structured light visual sensor is fully illustrated) are presented in the way that the principal foundation is set up to get the optimum structure design and simplify the calibration of the visual sensors.3. The conceptions of "feature point" and "feature line" are creatively established. As we know, "edge" has the definite meaning only in the real world where it refers to the shape of an object. When the object is imaged, its edge expresses itself in a certain given distribution of the gray on the image plane, therefore, it is hard to locate the edge on the image plane. The "feature point" and "feature line" are uniquely decided by the gray distribution and the corresponding recovering algorithms, thus, the ambiguity of the "edge" is avoided and exact meaning is given to the visual measurement and recovering algorithms, moreover, theoretical foundation is established to develop new algorithms for visual measurement and testing.In addition, the basic principles to recover the "feature point" and "feature line" as well as the criterion to assess them are discussed in detail.4. Several new algorithms of recovering the "feature point" and "feature line" at sub pixel accuracy are creatively presented. The results of the experiments show that all these algorithms work satisfactorily.5. A new way to calculate the distance of the parallel lines using the parallel lines least-square fitting method is creatively presented in the manner that the criterion of detecting the shape and position error is complied with, the deviation caused by the linear error and the parallel error of the lines is minimized. The mentioned method is experimented with the precision of 0.05 pixel.