| Abstract [eng] |
The final paper examines possibilities of standard industrial camera to determine the quality of details using image recognition software. The object of research is an automatic image recognition system used to determine the quality of details. The aim of this work is to investigate the automatic image recognition system used to determine the quality of details. Tasks - to analyze scientific and technical literature related to image recognition and methods, perform analysis of industrial image recognition solutions and camera calibration methods, describe the influence of different spectral light on image recognition, present the method of thread defect detection using a camera, describe research results. The analytical part examines the types of video cameras, lighting methods and camera calibration methods. The project part presents the physical and software required for research: camera, filters, light sources. The research part presents the research: – examination with a transparent filter and internal camera illumination, with illumination from the side and illumination from all four sides after elimination of natural light; – examination using a dome and internal camera lighting; – examination with red filter and internal camera illumination, with side illumination and illumination from all four sides after elimination of natural light; – examination with green filter and internal camera illumination, with side illumination and illumination from all four sides after elimination of natural light; – examination with blue filter and internal camera illumination, with side illumination and illumination from all four sides after elimination of natural light; – study with ND filter and with side illumination and illumination from all four sides after elimination of natural light; – the study examines the part by turning it at an angle. Conclusions and research results: 1. An analysis of the scientific and technical literature related to image recognition and methods has shown that image recognition solutions are relevant in industrial systems as a tool to achieve product quality and production efficiency. 2. The Zhang method (most suitable for the PIM60) was selected from three of the most commonly used calibration methods (Tsai, Zhang, and direct linear) for image recognition cameras. 3. After analyzing the scientific and technical literature, it has been established that when using image recognition systems to investigate surface defects, it is most reliable to use lighting in a dark field or bright field. Since shiny steel details were studied, the theoretically most reliable lighting methods were not suitable for the research. The study used simplified lighting methods using LED 5500º K light: lighting of the part from one side, external lighting from four sides, and lighting from above with integrated camera lighting. Lighting from one side worked best. 4. The methodology for detecting a thread defect in a part using a video camera is, in general, the introduction of a comparison provision (marking the properties of a quality part under the same conditions as the properties of other parts will be examined). With the help of various filters and lighting methods, the most suitable, high-quality properties detection conditions are sought, under which the object under investigation is best focused. Once the setting is made, the details under investigation are compared to it. 5. After the tests, the highest quality test results were obtained by observing the part at a distance of 90–100 mm, using a green camera filter and illuminating the part from one side. It has been found that by observing the part in approximately the same position (± 5 mm and ± 5°), as much as 100 % details properties can be recognized (quality parts meet specifications ≥90 and defective parts meet specifications <90). |
