Computer Vision News - March 2020

2 Summary RSIP Vision Projects 8 with Barel Levy Robotic surgery involves tiny incisions to operate on vital organs. This leaves little room for error and requires extreme precision. Before, many devices relied on magnetic or radio-based external sensors to calculate the endoscope location in the patient’s body. But this approach can lead to errors and rely solely on expensive devices with internal sensors. A recent breakthrough in 3D localization provides a cost-effective, highly accurate technique using inputs as simple as a video stream. The family of algorithms, called Simultaneous Localization and Mapping (SLAM) , creates a 3D map of a scene in real-time using a camera. Then the system calculates the location of the camera with extremely high precision. How Does SLAM in Endoscopy Work? Barel Levy , an algorithm developer at RSIP Vision , explains how SLAM calculates a real-time estimation of 3D TheReal-LifeBenefits of Real-Time SLAM in Endoscopy Applications Real-Time SLAM in endoscopy provides a cost-effective, highly accurate technique, that relies solely on inputs as simple as a video stream. models using the input of a handheld camera: “The algorithm locates key interest points in the images within the video stream. Next, it calculates speci f ic points of the image and unique features using methods l ike SIFT or ORB. The algorithm then matches new key points for each frame with the already detected keypoints. The algorithm optimizes the information by comparing the camera arrangements and key points in a 3D space to f ind the correct location in al l frames.” SLAM has other variations of the algorithm. One version uses a cloud of keypoints to create a 3D mesh depicting the video feed with high accuracy. From here, the device can use key points to triangulate the location of the new frame. Other variations arrange each pixel in the 3D model , rather than just the key points. Another version uses a pre-processed image rather than the original . Endoscopy,