Performance evaluation of depth map generation algorithm for stereo endoscopic camera

Abstract

This paper introduces a performance evaluation method for algorithms that generates a depth map using an image from a stereo endoscopic camera for image processing of laparoscope operations. The depth image was created by using a space-time stereo method by illuminating various patterns on scenes consisting of models of the 3D-printed organ model and actual organs from a pig, and the ground truth image was generated for each sub-pixel unit to achieve high accuracy and high precision. Different algorithms were evaluated using the ground truth image data. The number of effective depth pixels compared to the ground truth and the distance error was measured from algorithms based on an edge-preserving filter as real-time algorithms and quasi-dense algorithms. This paper presents an analysis of each algorithm from its evaluation indices to determine which algorithm is appropriate to compute the depth map from laparoscopic images.