Development of a surgical stereo endoscopic image dataset for validating 3D stereo reconstruction algorithms

In the last decades, endoscopic stereo images have been exploited to retrieve tissue surface information of the surgical site using 3D reconstruction algorithms. The application of such algorithms in Computer Assisted Surgery (CAS) tools for Minimally Invasive Surgery (MIS) requires a robust validation process in order to guarantee reliability and safety. 3D reconstruction algorithms are commonly evaluated comparing their result with respect to a reference Ground Truth (GT). However, few datasets providing endoscopic images and GT are openly available. Considering the increasing necessity of surgical datasets, the aim of this work is the generation of an Endoscopic Abdominal Stereo (EndoAbS) dataset composed of stereo-images with associated GT for 3D stereo-reconstruction algorithm validation. To recreate the surgical scenario, a polyurethane surgical phantom abdomen was built. Images were captured with a stereo-endoscope, while for acquiring the GT a laser scanner (calibrated with respect to the stereoendoscope) was used. This dataset is openly available on-line for the benefit of the CAS community.