Basic Geometric Shapes Detection Algorithm And The Application

In the process of image understanding and recognition, valuable information exists in the shape of object. Image-based detection technology, as an important part of automatic detection, is one of the basic research branches in the field of computer vision. The basic geometrical attributes of object is one of the main features in the image, so accurate detection to them has cardinal significance. Straight line, circle, ellipse is common geometric shapes in the nature or regular items. They are also key components of more complex shapes, which lay the footstone of subsequent advanced computer vision processing. They are widely used in the remote sensing, geographic information systems, face recognition, computer-aided design, target tracking, object positioning and traffic sign detection. That is to say stable and reliable geometry detection algorithm not only has cardinal significance for theory research but also has great market value.This paper studies the detection algorithms of the three basic geometric shapes of line, circle and ellipse mainly on the speed, reliability and stability. On the basis of the analysis of Hough transform (HT), randomized Hough transform (RHT) and other commonly used detection methods, it explores further improvements with theoretical research and experimental analysis. Main study work of this paper can be summarized as follows:1It proposes a novel sampling strategy in the process of random sampling when using RHT to detect line. Firstly, it determines the main direction of potential line according to the statistical gradient direction distribution of whole edge points. Then these edge points are constrained to small sampling range to reduce the invalid sampling. In order to build the detected line successfully it also records the starting and ending points of lines whenever necessary. Their reliability and effectiveness are confirmed by experimental results.2It improves the randomized circle detection algorithm. It constrains the sampling condition in the process of randomized sampling by setting threshold that is able to prevent from invalid sampling. In the decision of whether it is a circle, it establishes ring D and ovality O_r to reduce the bias and repair deformed circle. The experimental data proves the effectiveness of the algorithm.3. In order to resolve the problem of different lighting condition and the effect of complex nature background, it proposes a new detection algorithm of circular traffic signs. Combining the advantage of color segmentation of HSV color space and improved randomized circle detection, it locates the traffic sign quickly and accurately. The experimental results demonstrate the market value of this position method.