Tracking Feature Points in Uncalibrated Images with Radial Distortion

The appearance of moving features in the field-of-view (FoV) of the camera may substantially change due to different camera poses. Typical solutions for tracking image points involve the assumption of an image motion model and the estimation of the motion parameters using image alignment techniques. While for conventional cameras this suffices, the radial distortion that arises in cameras with wide FoV lenses makes the standard motion models inaccurate. In this paper, we propose a set of motion models that implicitly encompass the distortion effect arising in this type of imaging devices. The proposed motion models are included in a standard image alignment framework for performing feature tracking in cameras presenting significant distortion. Consolidation experiments in repeatability and structure-from-motion scenarios show that the proposed RD-KLT trackers significantly improve the tracking performance in images presenting radial distortion, with minimal computational overhead when compared with a state-of-the-art KLT tracker.