Computer Vision News 22 Best Paper 3DV Hon. Mention Songyou Peng (left) is a Senior Researcher/Postdoc, and Zihan Zhu (right) is a Direct Doctorate student at ETH Zurich. Fresh from winning a Best Paper Honourable Mention award at 3DV 2024, they speak to us about NICER-SLAM, an extension of their previous work, NICESLAM, which is looking to overcome the limitations of traditional SLAM systems. Classic SLAM systems focus on accurate camera tracking results but often struggle with accurate mapping due to their reliance on sparse point clouds. By incorporating Neural Radiance Fields (NeRF), Songyou and Zihan aim to achieve precise camera tracking combined with robust surface reconstruction and enhanced color modeling for applications such as novel view synthesis. This NeRF-based approach, however, would usually depend heavily on depth sensors, which are not as universally accessible or usable in any scenario as RGB sensors. Depth technologies like Microsoft’s Kinect and Intel RealSense are expensive, and they can face challenges when capturing information in certain lighting conditions during outdoor scenes. NICER-SLAM represents an extension of the pair’s previous influential work, NICE-SLAM, but with a significant twist – it removes the depth sensor. This shift to monocular or RGB SLAM (represented by the extra ‘R’) opens up the technology to a broader audience, including those NICER-SLAM: Neural Implicit Scene Encoding for RGB SLAM
RkJQdWJsaXNoZXIy NTc3NzU=