Expanding The Field Of View For Endoscope Via Grin Lens

Grin lenses are widely used in miniaturized optical systems because of their advantages of light weight,small size and being assembled easily.When the grin lens is applied to the endoscope system,we can minimize the volume of the system to the greatest extent.The view angle of the grin lens commonly used in the endoscope is about 60°,while the view angle of the endoscope with the conventional lens set(within 4mm in diameter)can reach 80°,so its performance for field of view is not enough to replace the conventional endoscope completely.For the problem of narrow view for grin lens endoscopic,we designed and built a multi-ocular endoscope system based on grin lens.Then we collected experimental sample images and established a set of algorithms,including feature point detection algorithm,image fusion algorithm and annular image stitching algorithm,to expand the field of view of the system.The research results of this paper are as follows:(1)We used Solid Works software to design the multi-ocular structure,and realized materialization by 3D printing.The system was built after selecting the appropriate optical components such as grin lens,relay lens camera and so on.The samples were observed to record the endoscopic images after finishing the above work.(2)In view of the barrel distortion caused by the grin lens,the polynomial distortion correction model is used to correct the distortion,and the distortion correction effect is good,which is helpful to reduce the image registration error caused by the distortion.For the problem of black area around fisheye image,the effective area of endoscopic image is obtained by using continuous pixel detection method.And we also realized the detection of pixel value with this method.(3)For the problems that Harris feature points is uneven and its number varies significantly with the change of the picture.We established an improved corner response function and carried out two threshold measurement tests to achieve a relatively stable number of feature points under different kind of samples.When the detection of feature points was completed,the adaptive non-maximum suppression method was used to screen the feature points,which can make the points to be accurately and uniformly distributed.We used nearest neighbor ratio method based on Euclidean distance to obtain the optimal matching point pairs while reducing the calculation amount.Finally,the spatial transformation matrix of the image was obtained by using the random sampling consistent algorithm.(4)We carried out several image fusion experiments with different algorithms,including fading in and out algorithm,best suture line algorithm,and pixel gray value maximization algorithm.Using information entropy,energy gradient and standard difference to evaluate the fusion results,the fusion method of maximizing pixel gray value was finally chosen.For the problem of the matching error and ghosting brought by accumulative deviation when using conventional method,we presented a novel image mosaic strategy based on ring multi-ocular images.This method takes advantage of the same area of the central image and the surroundings to increase the common area between the fused images,which can realize the more accurate fusion of multi-ocular images and reduce the phenomenon of ghosting.The accurate image with wide field of view will be obtained by the method.Finally,the quantitative analysis was carried out to analyze expansion ability of the system in field of view under different object distances.The best working distance of the system is about 4mm.Compared with the monocular system,the area of the view is expanded by more than twice,and the equivalent field angle of the system is expanded to 78.7° in this case.