next up previous
Next: The Estimation Algorithm Up: Localization with Automatic Feature Previous: Horizon Detection

Mountain Detection

Mountain peaks are detected by analyzing the horizon generated from the image. An A*-type search, similar to the search described by Thompson [14], starts from every line that satisfies some minimal requirements of size and orientation. The line is glued to other nearby lines that could form the image of a mountain; large, horizontal structures are favored by the search. If the search procedure finds indication that a mountain has finished (no nearby lines could be followed), the highest point of the mountain is detected and taken as the peak. A bearing is then obtained as the angle between the center of the image (determined by the camera calibration parameters) and the position of the peak.

The mountain detector must be able to distinguish small structures on a nearby mountain, such as trees or rocks, from distant mountains. This is done through an analysis of the smoothness of the horizon boundary: if the boundary is smooth in a given area, and there is indication of a mountain, then a distant mountain must be acknowledged; if the boundary is too rough, then a distant mountain is not possible and the detector looks for near and large mountains.

The bottom window in Figure 2 presents real results of the mountain detection system. Instead of looking at individual images, we show results for a mosaic of five images. Figure 4 shows results for data provided available from Dr. Thompson [14]; the images are large mosaics generated in Dromedary Peak, Utah.



Figure 4: Images from Dromedary Peak

Fabio G. Cozman