Co-Fusion
lastmod 2026-11-10created 2020-06-152 edits
Co-Fusion is a dense SLAM system that separates multiple moving objects in RGB-D camera video in real time, tracking each one while reconstructing it as a 3D model. Whereas traditional SLAM assumes a static scene, it segments objects from the background using motion or semantic information and maintains a separate surface model for each object. It is used for mapping dynamic environments, such as scenes where a robot manipulates objects.
Real-time Segmentation, Tracking and Fusion of Multiple Objects