Research On Circular Object Detection Algorithm

Circular objects such as oiltank and radome have always been the key targets in military system,structural components of satellite such as solar panel and antenna are presented as circular profile. The circular object is not only an important visual target subset and can be seen everywhere in our lives, but also presented in a variety of industrial applications. As a basic technical solution, research on circular object detection algorithm has a great influence on national defense, national economy and social development.After doing research and analysis, this paper proposed three solutions for object detection to solve the drawbacks of background interferences, blurred edges and small scale. They are curve fitting method, feature extraction method and model designing method.Analyzing the traditional circular type object detection algorithms such as the Hough transformation, they cost much space and time, this paper proposes a fast detection algorithm based on a circular arc grouping fitting. This algorithm takes fully advantage of the edge information of the target, according to the gradient direction and convex characteristic of the arc edges of objects, it groups the arcs. Finally the algorithm fits the parameters and the circular object is detected through pole and polar line property and Aguado Theorem.For boundaries blurred images, because it can't get enough edge information, edges based algorithms fail. In this situation, we propose a method by extracting object features. We extract HOG and LBP features first, then train features with support vector machine and get the classifier parameters. Finally objects with blurred edges can be detected.Considering the fact that some circular targets are too small, this paper detects this kind of object by constructing structural model based on structural information. The algorithm takes advantage of circular shape and the structural characteristics of the object and designs a detection model. It dynamically updates the model based on changes in the shape and scale. The algorithm enjoys good adaptability on small scale targets. Finally, model based object detection algorithm has been transplanted to run on a DSP chip. We do the algorithm optimization both in time and space.