Research On Parts Recognition And Measurement System Based On Machine Vision

As the development of modern manufacturing technology, the model identification and sizemeasurement technologies is required to develop in the direction of network, high speed and highprecision. The traditional inspection methods can’t meet the special need of modem manufacturing.Industry factories begin turning their attention to machine vision-based detection automation andassembly automation, after solving the problem of production automation. Machine vision detectiontechnology has the advantages of non-contact, high precision and rapidity. The non-contactidentification and measuring methods based on machine vision has become an important developmentdirection of precise measurement technology, machine vision detection technology has become one ofinterest topics in industry.In this paper, according to the parts recognition and measurement requirements of actual factoryproduction, we studyed the following several aspects of content:（1）We analyzed the research status of parts recognition and measurement technology basedmachine vision at home and abroad.According to the parts recognition and measurement requirements,we design a general planning and completed the hardware form, software design and communicationdebugging. Eventually we formed a set of parts recognition and measurement system based onmachine vision.（2）We studied the method of contrast enhancement, filtering, binaryzation, edge detection,edge morphological processing for parts image preprocessing and so on. We analyzed the basicprinciples of various processing methods, and provided various of the image processing effects ofparts,which was in preparation for the subsequent parts model identification and size measurement.（3）We studied the method of parts image recognition. According to the requirements ofrecognition, we chosed the high reliability of recognition algorithm based on gray level and designeda general identification scheme. For the large amount of calculation which is the shortcoming of thealgorithm, we performed two-step pre-classifying on the image of the parts according to the insideand outside diameter of the parts and delimited the matching area to improve the matching efficiency.The location of the recognized parts is not fixed and has certain rotation in the actual application, sowe rotated the template image to produce different direction of template to match during theprocedure of recognition. We described the methods of production of parts image template productionand parts image recognition algorithm, gave the steps of recognition of parts image matching and the results of image matching, also analyzed the results.（4）We studied the calibration principle of linear and nonlinear parts measurement systemmodel and the measurement experiment. According to the calibration to determine the correspondingrelations between the image plane coordinate system and the world coordinate system. Combinedwith the corresponding relation between the image plane coordinate system and the computer imagecoordinate system obtained the corresponding relationship between the parts pixel coordinates in thecomputer image coordinate system and the parts physical coordinates in world coordinate system, thusobtained the actual physical size of the parts. We described the calibration method, contrast before andafter of the calibration results. For parts size were measured in the experiment using the calibratedsystem,and the measurement results were analyzed.