Research On Autofocus Method Of Remote Optical Fiber Metrology Camera

The Large Sky Area Multi-Object Fiber Spectroscopy Telescope(LAMOST)is the astronomical telescope with the highest spectral acquisition rate in the world,which largely depends on the high-precision and stable operation of thousands of fiber positioning units on the focal plane.In order to accurately convert the pixel coordinates obtained from visual measurements into spatial coordinate systems,fiducial fiber for calibration was designed.A specialized structure designed with fiducial fibers is used to install the target ball,so that its spatial coordinates can be measured using a laser tracker.The special uniform light structure inside ensures uniform brightness of the light spot on the end face of the fiducial fiber,and its precise center pixel coordinate can be measured through the light center of gravity method and circle center fitting.Using fiducial fibers data for polynomial calibration,subsequent work can directly bring pixel coordinates into the polynomial to obtain the corresponding spatial coordinates.LAMOST optical fiber positioning closed-loop control system uses cameras to measure the operating status of the unit in real time.In the process of visual measurement,the movement position of the unit fiber can be accurately obtained by using the light center of gravity method,which requires the spot energy to conform to the Gaussian distribution.In order to ensure the availability and accuracy of the optical center of gravity method,it is necessary to design an autofocus system suitable for LAMOST optical fiber positioning closed-loop control scenarios.It is difficult for traditional autofocus algorithms to evaluate the clarity of fibers and ceramic ferrules,which account for a very small proportion of the image.In order to meet the special requirement of autofocus on fiber targets under large-scale conditions,we propose autofocus judgment methods applied to LAMOST optical fiber positioning closed-loop control system under front and back lighting conditions.Under front lighting conditions,in order to greatly reduce the calculation time,the system first preidentifies the focus target through Faster RCNN,and then uses the optimized contrast algorithm to evaluate the image clarity of the corrected ROI.This thesis compared the algorithm with the Tenengrad gradient calculation and Laplacian gradient calculation using Sobel and Laplacian operators in OpenCV.The results show that the method only took one ninth of the latter’s time,but could achieve the same accuracy.Under back lighting conditions,we used the average number of spot pixels and the average brightness of spot pixels as the evaluation criteria for image clarity.This method can evaluate the image clarity during the process of identifying light spots,and also provide initial data for subsequent fiber positioning work under closed-loop control.This method can achieve excellent efficiency while ensuring accuracy.This thesis also studies the difference between front-illuminated and backilluminated focusing,confirming the existence of this difference.In order to analyze the impact of this difference on the actual fiber position detection,we used a laser interferometer as a reference,designed a set of experiments to compare the positioning accuracy of front-illumination and back-illumination focusing.It is concluded that the accuracy of back-illuminated positioning is much higher than that of front-illuminated positioning,but the front-illuminated focusing and back-illuminated focusing have little effect on the positioning accuracy.The autofocus method and research conducted in this thesis not only play an important role in the optical fiber positioning closed-loop control system of LAMOST,but also apply to the focus judgment and process design of the closed-loop control system of other multi-objective fiber spectroscopic telescopes,and have great significance for practical application.