MICCAI 2022 Daily – Monday

In this paper, Jingwei proposes a CPU-level real-time stereo matching method for stereoscopes , such as stereo endoscopes or stereo laparoscopes. Engineers can deploy the algorithm on their CPU hardware and achieve real-time 3D reconstruction. The benefits compared to using an algorithm based on a GPU are that you can save the computation of GPUs for other tasks like diagnosis, segmentation, localization, or registration. However, there are some problems to overcome first. “ The surgical scenario suffers from bad illumination, textureless surface, and dark regions , ” he points out. “ To solve this, I devised an innovative Bayesian algorithm to measure the uncertainty of the matching results . If the results are bad, I delete them or use some of them to do fusion and have a better, more robust reconstructed shape. ” When performing the stereo matching, the basic idea is to find the illumination consistency, so the pixel on the left image should have the same illumination as its correspondence on the right image. However, due to bad illumination, the illumination consistency can be dense. Also, if there are textureless surfaces, the left pixel can be registered to any pixel on the right side because all pixels share the same illumination. Jingwei Song completed his PhD at the University of Technology Sydney before pursuing a postdoc at the University of Michigan’s Robotics Institute. He just enrolled in United Imaging in China as a Senior Software Engineer and continues to pursue research in its Research Institute. He speaks to us ahead of his oral presentation this afternoon, which explores CPU- level real-time stereo matching for surgical images . Bayesian Dense Inverse Searching Algorithm for Real- Time Stereo Matching in Minimally Invasive Surgery 4 DAILY MICCAI Monday Oral Presentation