Li, Daqing (1998) Road sign recognition. Masters thesis, Concordia University.
A method for recognizing road signs in street scene images is proposed. We first developed a fast region growing segmentation method, combined with a 1NN smoothing filter, to work on color and produce robust segmentation results even when the image is noisy. An effective color distance measurement is developed to work with the segmentation. We then designed a hierarchical feature representation scheme, which organizes model features in a tree structure. The tree nodes are components that form the basic parts of a model, and the tree leaves are primitive physical features that can be mapped directly to basic features in segmented image. A matching error model was defined to work with the feature representation scheme. To deal with the complexity in object detection, we designed an anchor feature and its detection scheme for the use of alignment detection. Alignment transformations are further verified by a unified matching scheme, which integrates feature ordering and performance tactics into one process. We used a modified Hausdorff distance as the final verification method in matching a candidate to the model. An efficient algorithm in calculating the distance using Voronoi diagram based on edge map as point set is developed.
|Divisions:||Concordia University > Faculty of Engineering and Computer Science > Computer Science and Software Engineering|
|Item Type:||Thesis (Masters)|
|Pagination:||x, 88 leaves : ill. (some col.) ; 29 cm.|
|Degree Name:||Theses (M.Comp.Sc.)|
|Program:||Dept. of Computer Science|
|Thesis Supervisor(s):||Bui, Tien D.|
|Deposited By:||Concordia University Libraries|
|Deposited On:||27 Aug 2009 17:13|
|Last Modified:||04 Nov 2016 18:00|
Repository Staff Only: item control page