Integrated Object Models for Robust Visual Tracking
2009
The robustness of visual tracking, or following the movement of objects in images, can be improved with an explicit model for the objects being tracked. In this paper, we investigate the use of an object model in this way. An object geometric model can tell us what feature movements to expect, and what those movements reveal about object motion. We characterize the tracking problem as one of parameter estimation from incomplete feature tracking data, and apply the Extended Kalman Filtering algorithm to the situation. Having an object model integrated into the tracking system overconstrains feature trackers, so that erroneous tracking results are selectively ignored and feasible tracking results are used to optimally update the object configuration estimate.