Detection of triangular and rectangular objects in digital images

In this dissertation, novel methods are proposed for the detection of rotated triangular and rectangular objects in digital images. The proposed methods utilize recently developed and successful edge detection algorithm, and consist of detection and validation stages. In the detection stage, the proposed methods use line segments and construct triangular and rectangular shapes from those segments. The line segments detected by using edge detection algorithm are converted into line pairs according to their angles and distance between each two lines. The candidate line pairs are first combined with each other. If the triangular or rectangular shapes are not constructed, for triangular shapes these line pairs are combined with a single line segment, for rectangular shapes two line pairs combined, and then these pairs are combined with a single line segment by following the appropriate criteria. Finally, in the validation stage, the candidate triangles and rectangles are validated using Helmholtz principle and Number of False Alarms (NFA) computation. According to the results of the experimental studies, the proposed methods offer higher detection performances than Open Source Computer Vision (OpenCV) triangle and rectangle detection algorithms which are commonly used in computer vision field. Keywords: Geometrical Shape Detection, Triangular Object Detection, Rectangular Object Detection, Shape Analysis, Hough Transform