Computer Vision News - December 2022

46 Congrats, Doctor! Endoscopes have changed how we diagnose diseases. They allow us to examine the inside of the body without requiring large and painful incisions and the resulting recovery. But endoscopies require extensive training because navigation within the body is challenging for both humans and machines. Animal tissue has different properties than outdoor or urban environments. It is deformable, reflective, and lacks robust features. There is also no ambient light, which means the appearance of any given surface changes whenever the camera and its attached light source move. So, given all these challenges, how can computer-aided systems help clinicians or patients during endoscopy? During my PhD, I helped develop a system that promises to one day provide clinicians with a real-time 3D map of the humancolon, so they canbetter navigate within it. Such a system will allow its operators to plan their next movements or evaluate past performance. Thanks to learning algorithms, researchers can easily predict local 3D structures in urban environments. But due to the extreme deformation within the body (bowel movement, breathing, changes in body pose etc.), it is impossible to obtain ground truth Anita Rau recently finished her PhD with the Surgical Robot Vision group at University College London. Her research aimed to improve navigation during endoscopic procedures by estimating 3D structures from monocular video. She now serves as a Postdoctoral Scholar at Stanford University, where she will continue her research on 3D scene understanding. Congrats, Doctor Anita!!! Figure 1: Our synthetic data provides 3D information and camera poses.