北京理工大学 增强现实智能医疗实验室

Abstract—Although medical endoscopy has been widely used in surgical procedures in clinical practice, radial distortion induced by the small size of the spherical lens in the endoscopic images generally lead to depth perception error and wrong position correlation of human tissues. The paper proposes a novel method for radial distortion correction using spherical projection model. First, a series of images of the designed marker board at different imaging orientations are captured by the endoscope. Then, the corners in all images are detected and matched automatically. The spherical model thus can be constructed by fitting the feature points in the projection images. Finally, the relationships of the source image, corrected image and the projection model can be obtained by optimization of the matched features. The process of distortion correction is fully automatic that requires no any human intervention. Experimental results demonstrate that the distortion correction method is very effective, which can achieve a mean RMS error of 0.81 pixel.