Feature Point Based Polyp Tracking in Endoscopic Videos

There has been various research conducted in polyp detection in endoscopic videos but not much has been done in the field of polyp tracking in endoscopic videos which itself is a very challenging task because of the nature of endoscopic videos. This paper discusses a modified method of polyp detection and proposes a feature point based polyp tracking method which is first in this field. Once polyp is detected, with Affine-SIFT (ASIFT) feature points extraction and matching we locate it in the next frame. Then we compute the polyp´s image position with homography constraints and set up an interest window to accommodate it. In the tracking phase, we only focus on the interest window, detecting feature points from the window and updating the window´s position and size until the polyps goes out of the frame. Results shows that Affine-SIFT (ASIFT) being a fully affine invariant image comparison method does works really well in endoscopic videos compares to SIFT and SURF.