Mario Sznaier, Jennifer Dy
Date of Award
Master of Science
Department or Academic Unit
College of Engineering, Department of Electrical and Computer Engineering
computer engineering, electrical engineering, video data registration, Scale-Invariant Feature Transform (SIFT)
Electrical and Computer Engineering | Engineering
This thesis focuses on a dynamic-based video data registration, which can register a video sequence to another reference video, captured by a camera mounted on moving vehicles on the same road but at dierent time and date. Applications of the proposed system include road inspection, and accurate localization of vehicles on the road.
The whole procedure of registration is done frame by frame on the sensed and reference videos simultaneously. First, a Canny edge detection is applied on the road portion of the current frame. A modified Hough transform algorithm, which is improved by using orientation of gradients, is then introduced to search road lines from these edges. The temporal trajectory of each road line is modeled as the output of a dynamic system. A Hankel based rank minimization algorithm provides predictions of the road lines based on information from previous frames and guides a line tracking algorithm. Additionally, a Scale-Invariant Feature Transform (SIFT) is applied on the current frame. The SIFT features are matched between current frames of the sensed and reference videos to estimate the Fundamental matrix between the two views. The Fundamental matrix and line correspondences are used to nd point correspondences on the road plane between the two videos. In turn, these point correspondences are used to estimate the Homography matrix between the road planes. Finally, the video registration is accomplished by warping the sensed video sequence onto the reference video sequence.
Huang, Minyang, "Dynamic-based video data registration" (2011). Electrical and Computer Engineering Master's Theses. Paper 57. http://hdl.handle.net/2047/d20001056
Click button above to open, or right-click to save.