Detecting System of Weld Joint by Image Measurement

1994 Fiscal Year Final Research Report Summary

• Project/Area Number: 05452295
• Research Category: Grant-in-Aid for General Scientific Research (B)
• Allocation Type: Single-year Grants
• Research Field: Material processing/treatments
• Research Institution: Osaka University
• Principal Investigator: INOUE Katsunori Osaka University, Welding Research Institute, Professor, 溶接工学研究所, 教授 (90029067)
• Co-Investigator(Kenkyū-buntansha): OHKUBO Masashi Osaka University, Welding Research Institute, Assistant, 溶接工学研究所, 助手 (10233074) ; TAKAHASHI Yasuo Osaka University, Welding Research Institute, Assistant Professor, 溶接工学研究所, 助教授 (80144434)
• Project Period (FY): 1993 – 1994
• Keywords: Parallel Processing / Image Processing / Active Contour Extraction / Genetic Algorithm / Top Down Approach / Dominant Point Detection / Weld Joint Image / Neural Networks

Research Abstract

The fundamental works were done for the establishment of an automatic detecting system of the weld joint. The construction of high ability image processing and measuring system, the development of new algorithm for the object recognition using this system and the proposal of a new noise rejection method were performed for the above purpose. Their outline is shown as follows.
(1) Construction of Image Processing System of Parallel Connecting Type--The image data conversion and input system was constructed by rough connecting of parallel type data processing elements. The optimization of the co-operative parallel processing was examined on this system and the pattern extraction unit composed of the LSI of special made.
(2) The Development of New Algorithm for Extracting the Contour of an Object--The contour line of the object for detection was extracted by the algorithm so called 'Snakes'. The genetic algorithm (GA) was applied for solving the 'Snakes'. This method was compared with the pr eviously proposed methods. Furthermore, The 'BIGA (BiPhased Genetic Algorithm)', which was the improved version of the above GA,was proposed as new developed algorithm. It is proved that the effective extraction of the contour line is possible from an object in a noisy and dirty image.
(3) The Development of New Algorithm for Detection of the Dominant Points of an Object--A new method for the dominant points detection, which could be applied to the contour line extracted by the above described method, was proposed on the basis of the principle of top-down approach. The feature of the method is that the number of the dominant points is beforehand given based on a priori knowledge on the object. The superiority of this method to the ordinary methods was shown by the comparative experiment.
(4) The Proposal of the New Method for the Elimination of the Noise on an Image of the Weld Joint--The application of neural networks was proposed to eliminate noise due to arc light, etc., superimposed on the image of the weld joint, obtained by the light slice method.