Homography-based orientation estimation for capsule endoscope tracking

Wireless capsule endoscopy (WCE) is a developing imaging technology for screening the gastrointestinal system. Tracking of the swallowable capsule endoscope within the human body is typically performed by external sensor arrays enabling approximate localization and detection of its orientation. In this paper we propose a method for capsule orientation estimation based only on image features, without the requirement of any external equipment. The proposed method involves two steps: a) salient point detection and extraction of image features from these points, and b) detection of feature correspondences between consecutive frames and homography estimation. The capsule orientation is estimated by decomposition of the homography matrix into rotation and translation components. The results obtained by the application of this method on WCE video frames indicate its effectiveness and improved performance over the state of the art.