Object-Based Visual-Inertial Tracking for Additive Fabrication

In this letter, a system is presented to track the motion of a sensor-head relative to multiple objects of known geometry. Measurements from a monocular camera and an inertial measurement unit are probabilistically fused in a moving horizon estimator to obtain high-accuracy estimates. Methods for detecting tracking loss and automatically resuming tracking are presented. The performance of the system is shown through experiments tracking various objects and ground truth measurements demonstrate the system's ability to provide accurate real-time motion estimates. As an initial application of the system, a 100-brick structure with complex geometry was built by hand using the tracking system and an augmented reality visualizer to guide construction.