Tracking Moving Objects Improves Recognition.
Nimit Dhulekar, Andrew Felch, Richard Granger · 2010
We describe a new family of algorithms that analyze time-varying scenes, recognizing and tracking learned objects over time. The new methods are intended to address key questions of moving images, including unpredictable moment-to-moment changes in location, size, orientation, lighting, and occlusion. We introduce a novel task in which objects revolve and rotate while suspended from a mobile’s arms; the recognition & tracking algorithm incorporates characteristics of a number of prior published methods, combining them in a novel fashion to enable this newly introduced task. Other methods have found that improving recognition will improve tracking; we show that improved tracking improves object recognition. 1.