Automatic detection of isolated airstrips in satellite images

Abstract

Automatic Object Recognition in satellite images is a significant task in Computer Vision Applications. Airstrip detection in satellite images is a complex task due to several variations present in the capture process, object appearance, pose and resemblance to other commonly appearing objects i.e. roads, buildings. Previous research has inspected to recognize airstrips based on airstrip characteristics such as longest parallel straight lines, constant area grey value, length and width dimensions in aerial or satellite images at single resolution level. The analysis procedure that is applied only at a single scale may miss some information at other scales due to objects with varying features. Therefore, it is the best solution to carry out analysis at all scales. This research focuses on automatic detection of isolated airstrips in multiscale satellite images. Detection of airstrips in multiscale satellite images is a complex problem due to diversity in pixel information. High resolution provides the detailed information of airstrip along with many other linear objects, whereas in low resolution pixel information gets lost and airstrip seems to be a linear structure like lines or roads .Similarly at maximum altitudes airstrip seems to be linear structures which are clearly visible as parallel borders at minimum altitude. As in Google Satellite images, diversity in images scales is seen. It is in common use now a days. But besides its versatility, it is still an unsolved problem to detect airstrips in multiscale images.