Object tracking using log-polar transformation

In this thesis, we use log-polar transform to solve object tracking. Object tracking in video sequences is a fundamental problem in computer vision. Even though object tracking is being studied extensively, still some challenges need to be addressed, such as appearance variations, large scale and rotation variations, and occlusion. We implemented a novel tracking algorithm which works robustly in the presence of large scale changes, rotation, occlusion, illumination changes, perspective transformations and some appearance changes. Log-polar transformation is used to achieve robustness to scale and rotation. Our object tracking approach is based on template matching technique. Template matching is based on extracting an example image, template, of an object in first frame, and then finding the region which best suites this template in the subsequent frames. In template matching, we implemented a fixed template algorithm and a template update algorithm. In the fixed template algorithm we use same template for the entire image sequence, where as in the template update algorithm the template is updated according to the changes in object image. The fixed template algorithm is faster; the template update algorithm is more robust to appearance changes in the object being tracked. The proposed object tracking is highly robust to scale, rotation, illumination changes and occlusion with good implementation speed