Dense Mapping

Object-Based SLAM

Longterm localization and mapping requires the ability to detect when places are being revisited to “close loops” and mitigate odometry drift. The appearance-based approaches solve this problem by using visual descriptors to associate camera imagery. This method has proven remarkably successful, yet performance will always degrade with drastic changes in viewpoint or illumination. In this paper, we propose to leverage the recent results in dense RGB-D mapping to perform place recognition in the space of objects. We detect objects from the dense 3-D data using a novel feature descriptor generated using primitive kernels. These objects are then connected in a sparse graph which can be quickly searched for place matches. The developed algorithm allows for multi-floor or multi-session building-scale dense mapping and is invariant to viewpoint and illumination. We validate the approach on a number of real datasets collected with a handheld RGB-D camera.

Kinect Monte Carlo Localization

Efficient Scene Simulation for Robust Monte Carlo Localization using an RGB-D Camera. The approach makes use of a low fidelity a priori 3-D model of the area of operation composed of large planar segments, such as walls and ceilings, which are assumed to remain static.


A SLAM system capable of producing high quality globally consistent surface reconstructions over hundreds of metres in real-time with only a low-cost commodity RGB-D sensor. By using a fused volumetric surface reconstruction we achieve a much higher quality map over what would be achieved using raw RGB-D point clouds. Our system performs strongly in terms of trajectory estimation, map quality and computational performance in comparison to other state-of-the-art systems.