Computer Vision News - April 2022

43 Challenges in Videos for RAS surgeries present different challenges for the above applications , and ensuring proper function in real-time is another obstaclewhichneeds tobeovercome. “ This fieldhas important clinical implications, and as a computer vision specialist it presents interesting and fulfilling challenges,” Asher concludes. “It will definitely redefine the way surgeries are performed, and help patients get better results in non-invasive surgeries ”. Read More can be used to reconstruct an accurate 3D map of the full scene. Obtaining videos of surgical scenes paired with accurate ground-truth (GT) is time- consuming and expensive, so typically only a small amount is available. In order to train the CNN we need to supplement the ground-truth data using creative data augmentation methods, as well as synthetically generated data . Another significant feature which assists in RAS, is the ability to perform real-time measurements on the image from the video feed. Prior knowledge of camera calibration and tool dimensions is used to address this need. Using deep learning, key points on the tools are detected, and the image is calibrated based on the prior knowledge. This allows accurate 3D measurements within the field of view, even without stereo vision. Despite the advancements in this field, there is still a long way to go. Different Synthetic Data Real Data