Computer Vision News - April 2022

41 Challenges in Videos for RAS their background, this task is quite simple. Detecting and segmenting tools is feasible, but recognizing a specific tool requires previous knowledge, so the system must be trained on specific tools to allow that. Once the tools are recognized, their position needs to be calculated. This is essential when attempting to conduct tasks like suturing - the position of the suturing needle must be calculated with high accuracy. The best method to obtain absolute coordinates from a video feed is by using stereo imaging - two cameras recording simultaneously from different the phase is not associated with a single tool, ” Asher explains. Therefore, more advanced deep learning (DL) techniques were applied. A combination of convolutional neural networks (CNN) with long-short term memory (LSTM) networks which take into account the time dimension of the video. Results were significantly improved and the final algorithm can detect the different phases of the surgery successfully. The next step towards autonomous procedures requires tool identification . As the tools are remarkably different from Deep Learning result paired with Ground Truth (top vs bottom)