Abstract:

Road extraction from Remote Sensing imagery attracted great attention and with great difficulty. Direction of twisty road (winding road in mountainous areas, etc.) changes seriously, and has poor geometry consistency. Variable road width or interferences, such as shadow and shelter, etc, increase the extraction difficulty. To solve this problem, this paper proposes a tracking method for twisty roads based on improved Circular Projection Matching algorithm. First, this paper proposes two improvements based on standard Circular Projection Matching, including: (1) Proposes the approximation parameter for circular projection vector as the circular projection similarity control; An optimal radius is proposed in order to reduce the calculation of the Circular Projection Matching with enough road characteristics matained. (2) Presents an interference correction algorithm, to a certain extent, overcome the matching problem between selected template and target object due to difference of lighting condition, image noise and other factors. Secondly, proposes a road tracking method based on above improved algorithm, and also proposes an adaptive template matching mechanism. The matched road targets are proper weighted, and are considered as part of the followed template. This mechanism makes the template flexible and less affected by the variations and interferences of roads. Finally, SPOT-5 and GeoEye-1 Remote Sensing image is used for experiment. The experiment result shows that the proposed road tracking method is direction-independent, which is applicable to the twisty road extraction. Road centerline extraction error is about 2~5m, and road extraction accuracy is higher than 80%.

Key words: Circular Projection Matching, Twisty road, Remote Sensing imagery, road tracking, road extraction