摘要：This article introduces a new method for road and pivot tracks detection for high resolution pancromatic satellite images. The approach is divided on three main modules: firstly, a spatial derivative mask is applied to the image for extraction of magnitudes and directions of discontinuities, then a five level quantization is applied on the magnitudes and a hierarchical consistency algorithm is developed by checking and linking each window to its neighborhood prioritizing directions with higher and consistent magnitude values; secondly, the road stronger hypotheses are left while the weaker ones are cleaned from the images, with the purpose to eliminate disconnected pieces; and finally, a Hough transformation is applied on the road hypotheses in order to complete the roads detected and to also identify possible pivot tracks among them. The method was applied to images of the ALOS satellite of different areas of the west of state of Bahia, Brazil, and results are provided here.
