Robot Control via Region-based 3D Reconstruction

L. Markham, S.C. Melton, and Z. Dodds (USA)


Vision-based control, autonomous robotics, 3d reconstruction, modeling and visualization


Reconstructive visual control of mobile robots currently focuses on sparse image features, e.g., points and lines. This work extends the reach of 3d reasoning to inexpensive – and increasingly ubiquitous – platforms whose visual input consists of segmented regions, rather than raw pixels. We propose and evaluate a pair of algorithms that enable two fundamental facets of robot control: obstacle avoidance and landmark reacquisition. Our results suggest that space-carving outperforms triangulation when using image regions to reason about one’s 3d environment.

Important Links:

Go Back