Research On Connector Surface Defect Detection Algorithm Based On Machine Vision

The surface defect detection based on computer vision has been widely used in many fields.As an important part of automatic system, it can ensure the quality of the production.Traditionally, artificial visual and sampling methodswere used to detect the surface detection.However, these methods have a high error rate and affect the efficiency of production greatly. So, surface defect detection approach of some industrial parts has been studied in this paper. And an automatic detection system on the connector surface defects is developed.The main contents of this dissertation are as follows:1. The common defect types of connector are analyzed. A defect detection system for the connector combined with the structural characteristics of the connector image and machine vision technology is proposed.The system can be roughly divided into four parts, such as image acquisition, image pre-processing, image registration and defect recognition.2.Some image pre-processing methods are investigated and applied to the connector images.The existing pre-processing method are compared and the proper one is selected for the subsequent defect recognition.3.The improvement of the image registration method is one of the main contributions of this dissertation. Since the accuracy and speed of image registrationis important in the whole process,we investigate this part in depth. First, the common registration methods based on "Edge information", SIFT and SURF are implemented respectively and compared. When these methods are applied to the registration of connecter images, both of the accuracy and time consumption are not acceptable. Then we propose an improved approach to increase the speed and accuracy utilizing the special characteristics of the connector and SURF Algorithm.Firstly, the image is binarized by Otsu,and then morphological processed.Then the general outline of the connector is obtained by Canny edge detector. Secondly,Surf are employed to detect feature points. The proper pairs of feature points are selected out of the numerous candidates by the criteria of Euclidean distance and least squares method.Finally, the image registrationis realized with bilinear interpolation and the transformation matrix calculated by feature points.4.The defects of connector are classified and a corresponding detection method is proposed according to each type of defect.Image segmentation, differential, morphological processing, connected domain detection and defect area thresholding are used for plastic defects, e.g. plastic edges, broken plastic and more plastic. A local thresholding segmentation,morphological opening operation and closing operation process are used with the geometric characteristics for terminals defects such as pin missing, crooked terminals.