Measurement Of Shaft Diameters From Machine Vision

With the continuous development of science and technology, visual measurement technique has been the rapid development and wide application of. As a means of non-contact measurement, it has become increasingly attracted the attention by people. In this paper, computer vision-based measurement system technology and hardware and software constitute have been studied, sub-pixel edge detection,corner detection and camera calibration techniques have been in-depth analysised. Visual measurement was applied in the shaft detection system, and measurement accuracy is in line with the industrial measurement requirements.In the machinery industry, the shaft parts is an important component parts of machinery. With continuous improvement of parts quality, higher requirements were put forward on measurement of the parts of technology and the evaluation method. The traditional means of contact detection of shaft was slow and not suitable for online real-time detection and measurement. In this paper,a shaft measurement system was set up basing on visual inspection techniques and the geometric characteristics of shaft, through non-contact visual inspection, to improve the measurement of shaft speed, to achieve high-precision measurement of the shaft.In this paper, mainly basing on visual measurement technique, focusing on the key technologies of sub-pixel edge detection, corner detection, camera calibration technology and reverse the size, a shaft measuring system was established. Thesis work involved the following aspects:The sub-pixel edge detection algorithm basing on fitting method. The Application of visual inspection for measurement technology, First of all, need detect the image positioning of objects edge, can eventually reverse the geometric parameters of objects. Therefore, the image edge positioning accuracy of object detected will directly affect the final results. Early edge detection algorithm used for the pixel level, such as the commonly used Sobel operator, Roberts operator, Krisch operator, Prewitt operator, Laplacian operator, Gauss - Laplace operator and Canny operator and so on. Positioning accuracy of these algorithms was of the entire pixel, that is, to determine the location at the edge of a pixel with a pixel, but can not be broken down further. Sub-pixel edge detection techniques were first proposed by Hueckel, and existing sub-pixel edge detection method can be summed up in three types: basing on method of space-moments, based on the method of least squares (fitting) and based on the interpolation method. And the fitting method is fitting assuming the gray value of edge model of least square to obtain sub-pixel edge position, accessing to the edge of the sub-pixel accuracy is higher than the other two methods, and the method has a good noise robustness, stability. In this paper, based on the principle of least square method, using the edge of gray distribution model, a new sub-pixel edge detection method - sub-pixel edge detection algorithm based on Bessel edge model, and the procedure of the algorithm has been written.Experimental Analysis of the edge detection algorithm based on modified Bessel Model. Experiments have proved that the amendment of the Bessel model is necessary, and analysised of the fitting window size and gray-scale differences on the impact of the resolution of the edge detection algorithm. As sub-pixel detection algorithm of accuracy can be achieved one-tenth of a pixel, or even higher, it is difficult to evaluate the algorithm of accuracy. Due to the true position of the edge can not be determined, it is unable to determine that is it recent between the edge points were detected and the actual edge points. Therefore, we given a method for evaluating the accuracy-resolution of edge detection algorithm, and used the evaluation methods to compare with the previous sub-pixel edge detection algorithms.Corner detection. According to measurement accuracy, corner detection algorithm can be divided into two types of pixel-level and sub-pixel. One common pixel corner detection algorithm: Harris operator, Moravec operator and the Susan operator, and Forstner operator, some improvements Harris operator and the Bouguet algorithm provides sub-pixel accuracy for corner detection. Corresponding to chessboard calibration grid and grid-based board images, we gave a fitting on the edge of the sub-pixel corner detection method. According to Checkerboard-type grid for the calibration board, we firstly can fit four equation of sub-pixel edge, then get 4 points, respectively, and ultimately use the method of quality-force to determine the sub-pixel coordinates of the corner. The purpose of past the corner detection algorithm is not for high-precision measurement, resulting corner detection and edge detection to achieve sub-pixel accuracy using different mathematical methods. This will result in the same image feature detection, giving inconsistent results, and results in the benchmark of calibration and measurement different, affects the image measurement of accuracy. Therefore, as the criterion for the image measurement accuracy, we give a new evaluation method for corner detection accuracy, having used the evaluation method comparing the corner detection algorithm with the previous algorithm.Camera calibration. Camera calibration is the basis of measurement of using digital images. It sets up relationship between the camera position and the scene image pixel position, and the way is based on camera model, the known images coordinates of feature points and the world coordinates were used for solving the parameters of the camera model. At present, the commonly used methods include: Tsai calibration method, Weng's two-step calibration, Zhengyou Zhang's calibration method. Zhang's calibration method which can provide good initial values for solving speed and accuracy is relatively high, is a better calibration method. Based on the analysis of Zhang's calibration method, an improved calibration method is given. The method improves the distortion of the original calibration model, without lowering the accuracy of calibration, the calibration algorithm allows a more concise, easy to reverse by the image coordinates of the world coordinates, in line with the demand for high-precision image measurements.The use of CCD for non-contact measurement of shaft, the geometric information to be obtained is the length of shaft diameter. Unlike previous methods, this article is on the use of a single CCD camera to measure shaft. This paper presents a detection method forthe shaft, which consisted of the CCD camera calibration and sub-pixel edge detection method. This method based on modifying the distortion model of Zhang's calibration method, used of specially designed clamping fixed calibration board to calibrate external parameter of camera, to ensure that the gripping state of the measured object and the external calibration parameters is of the same. We use of fitting forms of sub-pixel edge detection algorithm to determine the two edges position of the shaft. Because of pinhole imaging theory, the space distance of two shaft edges detected in pixel plane is not shaft diameter(Pseudo-diameter), therefore the function relationship between pseudo-diameter and true diameter is given. According to the theory, combining of hardware equipment, a shaft measuring system is established.The innovative work of this paper are mainly the following:1. According to the theory of the point spread function and fitting, we first proposed a sub-pixel edge detection algorithm based on the modified Bessel model, improving the accuracy of sub-pixel edge positioning.2. We gave a method for evaluating the accuracy-resolution of edge detection algorithm, and used the evaluation methods to compare with the previous sub-pixel edge detection algorithms,proved that this method had a high positioning accuracy of sub-pixel edge3. We gave a fitting on the edge of the sub-pixel corner detection method. As the criterion for the image measurement accuracy, we give a new evaluation method for corner detection accuracy, having used the evaluation method comparing the corner detection algorithm with the previous algorithm, proved that this method is more in line with the needs of high-accuracy image measurement.4. An improved calibration method is given. Tthe calibration algorithm allows a more concise, easy to reverse by the image coordinates of the world coordinates, in line with the demand for high-precision image measurements.5. According to the theory and algorithms, combined with the geometric characteristics of the shaft, a image-based detection method for the diameter of shaft is given. Shaft diameter measurement system was established, and through different experimental measurement of shaft to verify the system has high accuracy.