Smart and Easy Object Tracking.
Petr Fejfar, David Obdržálek · 2014
Abstract. In this work, we present a system which is able to track objects over time even in situations the objects are not apart enough to get separated input data. The system processes distance data repeatedly acquired from a distance sensor. More specifically, it tracks balls rolling on the floor using laser rangefinder measurements. The noisy data is first filtered, then processed for ball detection and consecutively paired with long-term data of objects movement. Finally, Kalman filter helps to bridge the drop-outs of object position information caused by interactions between the balls and occlusions. Based on the tracked movements and smart predictions, the system is able to cope well with two principally different and poorly distinguishable situations: when two balls pass close to each other without touching and when two balls collide and bounce away. The system has been successfully implemented on a real robot equipped with a short-distance IR laser rangefinder. 1