| Summary: | 碩士 === 國立彰化師範大學 === 機電工程學系 === 101 === Industrial endoscopy has been widely used in the field of aviation, automobile repair, oil, plumbing and electrical technician, public security, precise instrument. The industrial endoscopy is small enough to enter the places that normal camera can't approach. Because of the narrow field of view, the operator only can get a part of whole information. This constrain results in non-comprehensive view so that the operator can't make a right judgment. Therefore developing a image stitching algorithm applied to industrial endoscopy is helpful.
In this paper, the endoscopy consist of a CMOS color camera with dimension of 5.5mm 15mm and four white LED. By use of video capture card, the camera connect to computer and then taking pictures through user interface. Then deriving the camera parameter and fixing the distortion image use a planar checkerboard. In the procedure of taking photos, the angle of view is variant so we use the SIFT (scale-invariant feature transform) that is invariant to image scale and rotation. The overlap region can be acquired by feature matching then estimate the 2D homography for image pair. Finally correlative image projects on the target image. There are discontinuous crafts caused by illumination difference at overlap region while stitching. Accordingly, using cross blending method makes the stitched picture more seamless and nature. After capturing images and stitching experiment, it is successful in stitching muti-valve images and circuit board images.
|
|---|
