Object detection in very high resolution satellite imagery based on Computer vision techniques

点击翻译按钮获取中文摘要

Huge amounts of data are acquired with either spaceborne or airborne sensors to serve as basis for cartographic data usually being drawn manually. Very high resolution (VHR) aerial and satellite images provide very valuable information. In particular, detecting objects (e.g.: building, cars, boats, etc.) from these images requires a specific consideration, since this information may be used in several remote sensing applications, such as automated map making, urban planning, and land use analysis, etc.. Unfortunately, it is tedious for a human expert to manually label objects in a given aerial or satellite image given the large geographic areas covered with many objects in them which means that analyzing manually the image may take time. Thus, map databases are often outdated. Moreover, in case of crisis situations evoked by, for example, earthquakes or inundations, rapid mapping is essential to organize instant response actions. Therefore, development of robust and fast building object algorithms on VHR aerial and satellite images has become a necessity. Independently, the detection, classification and segmentation of objects in images is a popular subject in computer vision. The goal of this proposal is to review existing methods for object detection in the Compute Vision research area and see their potential applicability to VHR satellite imagery.