Research On Recognition And Detection Of Complex Injection Molded Parts Based On Machine Vision

Injection molded parts have the feature of low cost, light weight, small transmission noise, and has been widely used in most kinds of instruments with high precision and micro-electro-mechanical products. Since the injection molded parts are easy to produce defects during the molding process, it is necessary to carry out the testing procedure to ensure the quality. At present, most domestic manufacturers still adopt the approach of human detection which is time-consuming and inaccurate. In order to improve the production efficiency and increase detection precision, machine vision is applied in this study to recognize and detect complex injection parts. As a kind of non-contact online detecting technology, machine vision has the advantages of fast speed, high precision and automation. It uses image acquisition equipment to convert object to image which is processed by the program to obtain the morphological information of object.In view of the shape and size characteristics of complex injection parts, research on the recognition and detection of complex injection parts based on machine vision was conducted in this study. The main research contents of this study are as follows:Based on the requirements of recognition and detection of complex injection parts, the overall scheme of system was designed, then completed hardware configuration and experimental platform. In order to realize the mutual transformation between the camera coordinate system and the world coordinate system, calibration of camera's internal and external parameters was realized by HALCON software based on the camera imaging model.After setup the hardware platform, the image preprocessing algorithms, including image filtering, threshold segmentation, morphological processing and edge detection, were studied to facilitate the follow-up work. Experiments about the effect of mean filter, median filter and Gauss filter were carried out, which show the median filter is best in the retention of details. Threshold segmentation was used to extract part area which may contain holes and noise which was eliminated by morphology algorithms. Aiming at the advantages and disadvantages of common edge detection algorithms, an optimized edge detection algorithm based on gradient operator was also proposed, which had the detection result of single pixel edge with less noise.Based on the results of image preprocessing, the recognition methods of parts of different types and sizes were studied, including the template matching and BP neural network identification. Grayscale template and edge template were made for template matching, and HALCON was used to complete the design and training of neural network. The recognition results show that the accuracy of the two methods is comparable, but the BP neural network based recognition method is obviously less time consuming.Detection research on parameters measurement of plastic gear was studied. Multiple methods including centroid method, addendum circle fitting and root circle fitting were used to obtain center coordinates. The average value of center coordinates calculated by these three methods is regarded as final center coordinate. The tooth number, circular pitch and diameters of addendum circle and root circle were measured to test whether the gear is qualified.Based on the above mentioned researches conducted in this study, visual recognition and detection system was developed by HALCON, OpenCV and C++Visual 2010. The system mainly realizes the functions of image acquisition, image recognition, gear size measurement and the results display. The software interface contains four windows of real-time video, recognition and detection results, image processing results and operation. The experimental results show that the system has a high accuracy of recognition and detection, which shows the value of practical application in the industry.