Proposal for Endoscope Robot with Surgical Scene Recognition using Image Processing

Abstract

A method to recognize surgical scenes using only image processing has been developed to control endoscope robot semi-automatically. Some basic methods of image processing are designed for the scene to make a triangle working area using forceps required for endoscopic surgery. The FAST (Features from Accelerated Segment Test) for the corner detection and the Delaunay diagram for the triangle automatic generation were applied to a still surgical image. The binarization and Hough transform using HSV model for the feature-point detection were applied to sill endoscopic images. The MoG (Mixture of Gaussian Distribution) and GMG (Godbehere, Matsukawa, Goldberg) for probabilistic background subtraction were applied to moving endoscopic images. The expected triangle was not shown, but the results indicated the feature design to allow robotically assisted endoscopic surgery.