This paper presents a technique for extraction of unpaved roads from high resolution satellite images. Sandy region candidates were obtained through thresholding of relevant colour space components from HSV and YCBCR. Logical combination of these candidates were performed to give the sandy region. Morphological operation (Dilation) was used to refine the sandy region, while Hough transform was used to extract road section from the whole sandy region. The developed algorithm was implemented in MatLab and tested with samples of satellite images acquired from Google Earth Map. Subjective (visual quality inspection) and objective parameters (Gap Statistics) were used for performance evaluation. The result shows that the new algorithm successfully extracts unpaved roads from satellite images. Areas of application includes security and surveillance systems, such as tracking of insurgents and kidnappers in tropical regions up to semi-arid regions.
Keywords: Colour Space Models, Dilation, Geographical Information System., Hough Transform, Unpaved Roads
