Oil Tanks Extraction from High Resolution Imagery Using a Directional and Weighted Hough Voting Method

Abstract

Automatic objects detection, such as oil tanks, is an important application of high resolution satellite image, which can be used in disaster prevention, environment monitoring, and some military tasks. Hough Transform is one of the most commonly used method for circle target detection in the literature. However, complicated details and background in high resolution image make it difficult to find a satisfying edge map, which limits the accuracy of oil tanks extracted by edge based Circular Hough Transform methods. In this paper, we propose a novel process to detect circles directly from the gradient image rather than the edge map. By this way, we avoid the problem of edge extraction. On the other hand, by searching for the most suitable circle center in a buffer region, rather than just selecting the local maximum, we improve the accuracy of circle center location significantly. We have applied this approach to different high resolution satellite images, and find both the completeness and the accuracy of the proposed algorithm superior to the traditional edge based Hough method.