Dense Depth Estimation for Surgical Endoscope Robot with Multi-Baseline Depth Map Fusion

Dense depth estimation in endoscopic images can provide surgeons with important information for performing accurate minimally invasive surgeries. However, it is difficult to estimate the absolute depth of the scene based on monocular endoscope. Depth values in endoscopic images change drastically during the operation, which make it hard to estimate them with a fixed baseline. In this paper, we propose a depth estimation scheme with multiple baselines. The monocular endoscope is moved horizontally by a robotic endoscope holder to generate stereo images. A pixel-level depth map fusion algorithm is designed to combine depth values estimated with different baselines. Experimental results show that the proposed method improves the accuracy of depth estimation and the visual quality of depth maps.