Reliable 2D Tracking using good texture and edge features for Robotic Vision

We present an algorithm for highly reliable tracking of planar objects using visual cues like texture and contour in presence of feature correspondence errors. These two cues are integrated using a probabilistic formulation. The integration is based on quality goodness factors. The goodness criterion is a generalization of the well known "good features to track" concept to the both point and edge cases. The motion model of the object is computed as a homography between reference and current frames. A probabilistic formulation of the problem is proposed and implemented using particle filters. Tracking for geometric computation is useful in applications like object grasping, 3D reconstruction, augmented reality, etc. The algorithm combines contour and texture information in a novel manner to achieve robustness that outperforms the state of the art methods, which is justified by the results of experiments