Realtime Registration-Based Tracking via Approximate Nearest Neighbour Search

We introduce a new 2D visual tracking algorithm that utilizes an approximate nearest neighbour search to estimate per-frame state updates. We experimentally demonstrate that the new algorithm is more efficient than the standard registration-based tracking algorithms and that it is capable of accurately estimating larger per-frame motions. Finally, the algorithm is demonstrated to be more robust in a vision-controlled robotic alignment task.