Smart Camera Network Localization Using a 3D Target

We propose a new method to localize in three dimensions the camera-equipped nodes in a smart camera network. Our method has both lower costs and fewer deployment constraints than a commonly used computer vision-based approach, which is to opportunistically determine feature points in the overlapping view of pairs of cameras, compute the essential matrix for all such pairs, then perform a bundle adjustment to both refine all camera positions and orientations a determine a common scale. Our method utilizes a feature point filled 3D localization target with efficient detection algorithm to determine the projection matrix for a camera viewing the target. Because the projection matrix gives the position and orientation of the camera in the external coordinate frame of the localization target, two or more nodes simultaneously localizing themselves to the target are automatically localized in the same coordinate frame. This technique can be used to localize a smart camera network with connected views because as the target moves through the network each node will localize itself to at least two target positions that are related by an easily determined rotation and translation and which can be used to globally align all node positions and orientations to any single network-viewable target position. We present results from a real indoor network and suitably designed localization target, and show that our method can accurately localize the network when the target’s feature points fill less 5% of the frame. Because the target can be relatively small in frame, pairwise camera overlap can also be small.