Computer Vision News - April 2022

42 Robotic Assisted Surgeries was reconstructed. Generalizing this method to all the pixels in the image for complete scene 3D reconstruction is difficult, since the anatomy, unlike the surgery tools, is very monotonous and lacks distinguishable key-point features. Therefore, deep learning architectures - that separately compute features for each of the frames and calculate the disparity of each pixel in the frame - were implemented. This in turn angles . Most robotic systems today come with stereo imaging so obtaining this data should not be problematic. “We were able to calculate position even from monocular imaging, however, stereo depth is much more accurate than monocular depth, ” Asher clarifies. By segmenting the tools in each image and calculating the disparity (=difference of the location in the 2 frames) of the segmentation, the 3D position of the tool Segmentation and Tool Classification Tool Segmentation