(A) feature-based planar object tracking method using dynamic feature update

Abstract

In this thesis, a feature based planar object tracking method and an effective model update method are proposed. Object tracking is one of the most attractive research areas of image processing and computer vision community, as multimedia infra advances. The proposed tracking method is motivated for extracting accurate license plate region of vehicle. Proposed method is designed to utilize the geometric composition of target area’s feature points, while achieving high tracking performance when the input sequence contains drastic geometric distortion or severe occlusion. The tracking method is operated as follows. First, local features of candidate image are extracted by SIFT. Secondly, the movement parameter modeled by homography is calculated by RANSAC. Finally, reference for the next tracking iteration is generated by dynamic feature update method. In the proposed dynamic feature update method, highly informative information is selected as much as possible for the next reference. Experiments on several real sequences with significant geometric distortion and severe occlusion show that proposed algorithm achieves high tracking performance.