Cognition System of Bolt Hole Using Template Matching

Abstract

ABSTRACT:In the construction field, the bolting robot provides convenience and reduces a process time of assembling steel frames. When assembling steel frames, the bolting robot finds a bolt hole with the help of camera. If the light conditions are either too bright or too dark, however, it is hard to find a bolt hole in the input image because the contrast between bolt hole and background is vague. In order to improve detection accuracy in such environment, this paper proposes the image processing algorithm using template matching (TM). First, the input image is converted from RGB to YCbCr in order to reduce influence of shadow. Then we separate Y channel of image in darkness or extract edges from Cb and Cr channel of image in brightness. Template image is created with the radius which can be calculated with the distance to a steel frame. We used TM to create the normalized cross correlation (NCC) image which shows correlation between processed image and template. Finally, bolt holes are extracted from the high-correlation part of NCC image. Experimental results show that the proposed method is robust to detect bolt holes under various illumination conditions.

Object's Centroid Localization Using Hu-Flusser's Moments Invariant

“…Normally, template matching (TM) is a standard method for locating image features. There are many developments involve this technique to state of the art about the normalized cross correlation (NCC) [4][5][6][7]. But for simplicity, we just use the standard correlation theorem [8] to be a test bed in this research.…”