Research On Precision Alignment Of Optical Fibers Based On Binocular Stereo Micro-vision

The alignment and fusion of the optical fiber plays a fundamental role in building thelarge-scale fiber-optic communication network. Up to date, the image-based method has beenone of the most prevailing approaches to actively align two optical fibers. This methodpossesses good features, such as high resolution, good robustness, and high alignmentprecision. Currently, the commercialized optical fiber alignment and fusion machinegenerally utilizes two light microscopes to help adjust the relative positions of the two opticalfibers at the axial and radial directions. Usually, it requires two very precision clampingstages to constrain the other four degrees of freedom (DOFs); therefore, the orientations ofthe two optical fibers can be held the same.The growing manufacturing technology of the industrial lenses gives birth to thetelecentric lenses which possess excellent performances in the resolution and lens distortion.Thus, telecentric lenses are broadly applied in the precision visual metrology. In this paper,we substitute the telectric lenses for the light microscopes, and present an alignment methodbased on the telecentric stereo micro-vision. The optical fibers can be adjusted in all6DOFs,which avoids the usage of the precision clamping stages. In general, the main works of thispaper can be categorized into the following three parts:First, the telecentric stereo micro-vision system needs to be calibrated. The purposes ofthe calibration are, on one hand, to acquire the one-to-one mapping relation between the3Dscene points and the2D image pixel points; on the other hand, to define a world coordinatesystem for the subsequent3D reconstruction and measurement. With respect to the generallysmall field-of-view (FOV) and depth-of-field (DOF) of the telecentric lenses, we present ahighly compatible, easy-to-use, and precise method for calibrating the telecentric stereomicro-vision system.Second, the feature points of the optical fibers, from the left and right images in whichthey are actually not very much evident and easily identifiable, need to be extracted and thenmatched. Since the two cameras are set up in different positions, so the corner features of theoptical fibers in the two images are not correspondent with each other. Therefore, we need tofind out some image features that do not change with the camera position. In this paper, we take use of the affine epipolar geometry and the geometry of the optical fiber, and present amethod that can successfully realize the image feature extraction and matching.Third, using the vison system, we complete the3D reconstruction and measurement ofthe optical fibers. According to the measurement results, the precision stage is driven to alignthe two optical fibers. In order to reduce the measurement errors and improve the alignmentprecision, we analyze the source of the errors and conclude that the errors mainly come fromthe calibration, image processing and feature points matching. Additionally, if the definedworld coordinate system does not coincide with the stage’s coordinate of motion, then anextra hand-eye calibration is required to work out the relation between the two coordinatesystems.