We present a deeply integrated method of exploiting low-cost gyroscopes toimprove general purpose feature tracking. Most previous methods use gyroscopesto initialize and bound the search for features. In contrast, we use them toregularize the tracking energy function so that they can directly assist in thetracking of ambiguous and poor-quality features. We demonstrate that our simpletechnique offers significant improvements in performance over conventionaltemplate-based tracking methods, and is in fact competitive with more complexand computationally expensive state-of-the-art trackers, but at a fraction ofthe computational cost. Additionally, we show that the practice of initializingtemplate-based feature trackers like KLT (Kanade-Lucas-Tomasi) usinggyro-predicted optical flow offers no advantage over using a carefuloptical-only initialization method, suggesting that some deeper level ofintegration, like the method we propose, is needed in order to realize agenuine improvement in tracking performance from these inertial sensors.
展开▼
