Evaluation of Feature Detector-Descriptor Using RANSAC for Visual Tracking

Abstract

Vision-based tracking is an essential prerequisite to a growing number of applications in computer vision. Any tracking algorithm is expected to deal with problems like a change in intensity, scale, pose, and camera motion. This paper summarizes the implementation of such algorithms stating their merits and demerits under various transformation and distortion of images like blurring, noise, intensity variation, and rotation. Also, implementation of an object recognition system is done that uses image matching techniques with RANSAC algorithm to identify the objects in a given scene using their 2-D images. AKAZE, BRISK, DAISY, FREAK, ORB, SIFT and, SURF algorithm has been used for feature detection, extraction, and matching. Then the outliers are removed by RANSAC algorithm and homography detects the object for each image matching algorithm.