An approach to automatic object tracking system by combination of SIFT and RANSAC with mean shift and KLT
Bhakti Baheti, Ujjwal Baid, Sanjay Nilkanth Talbar · 2016
Object recognition and tracking are important and challenging tasks in many computer vision applications. Difficulties in object recognition arise due to occlusion, clutter and geometric transformations present between pair of images or frames. Challenges in tracking include ability to deal with abrupt object motion, nonrigid object structures, change in appearance patterns of scene and object, occlusions present and camera motion. To deal with these challenges, we have explored the effectiveness of SIFT for feature extraction and RANSAC for homography estimation in object recognition. This makes system invariant to geometric transformations, illumination variations, partial occlusions and clutter. This automatic object recognition approach is used to automatically detect the object in first frame of video and then it is tracked in subsequent frames. Object tracking is implemented by Mean Shift Algorithm and KLT tracker. These algorithms have ability to handle partial occlusion and clutter. Combination of Mean shift and KLT with SIFT and RANSAC makes the system automatic.