Research On The Geometric Measurement Of Mechanical Parts Based On Image Processing Technology

Measurement plays an important role in the development of machinery industry. Traditional labor-intensive manual contact measurement costs large human and material resources , and has very low detection efficiency.Using digital image processing technology for geometric on-line measurement of mechanical parts on industrial production line has advantages of rapid speed and complete non-contact, and is one of the key means of ensuring the quality of the spare parts. The main content of the project research is the two-dimensional geometric quantities of the mechanical parts using the digital imaging measurement method. Mechanical parts are generally composed of straight lines and arcs, therefore, the detection of linear and circle is an important part of the image measurement of components. For the existence of problems and limitations of detection algorithms used in the actual linear and circular measurement of parts, a number of linear and multiple circle detection algorithms have been mainly studied in the application of machine parts measurement, including digital image preprocessing, edge detection and refinement, sub-pixel positioning, form and position error measurement, etc. The accurate measurement of the straight line parts and porous parts containning multiple basic geometric elements is completed.The main research content, work and production of this thesis are as follows.1. Introducing the multiple lines detection algorithm based on the thinking of "first local Hough transform, after the global least-squares fitting"; Solving the uncertainty problem of test results when detecting a straight line using the random Hough transform, as well as the the interference problems of Follow-up line detection when detecting many lines; It has good robustness, immunity to interference and flexibility.2. Two methods are used to solve contemporary detecting problems of parts with several circle elements in porous components measurement. The first method is to separate these circles, and then doing point Hough transform for each circle alone , which not only reduces the cumulative ineffective, but also removes the noise effectively; The second method is a universal multi-circle detection algorithm, which is nearly applicable in the multi-circle detection task in all cases,3. Researching an improved sub-pixel algorithm, and this algorithm is suitable for the image of real-time measurement tasks. The edge thinning algorithm of gray-scale image is introduced, which avoids losing the edge with smaller amplitude in the edge graph; Sobel templates with four directions are adapted to got more clear and complete single-pixel edge. Algorithm is simple but has a strong anti-jamming capability, and can locate the edge of the object Measured accurately.4. Following the size measurement of basic geometric elements, the shape and position errors measurement problems is researched based on the least square method, including straightness, parallelism, roundness, etc.