Revision history [back]

I guess you are talking about the point cloud, which only represents the measured points. For the reason you state people use probabilistic occupancy maps, which consider the implicit "free space" between perceived point and sensor. RGBDSLAM uses the octomap library to create such an occupancy map. In this map the removed object will (slowly) be erased, when you revisit the place where it has been. You can either save the point-cloud as octomap when you are done, or create it online.