Towards Automation of Information Extraction from Aerial and Satellite Images

Abstract

Since photogrammetry was developed more than 100 years ago as a technology for map production and the measurement of objects on images, photogrammetrists have attempted to improve the efficiency and accuracy of the process. These attempts commenced with the development of analogue approaches to solving the major computations in converting image coordinates on photographs to coordinates in a ground or object system. Subsequent developments included computer driven instruments, the ‘analytical stereoplotter’ and in the early 1990s, purely digital systems based on digital image acquisition and processing. The precision of all components of the photogrammetric process was continually improved so that smaller image scales could be taken to achieve the required accuracy on the object, thus improving efficiency and reducing costs of the mapping process. Digital image processing in photogrammetry currently enables the determination of elevations and the production of digital orthophotos more rapidly and with greater efficiency than could be achieved with analogue instruments. However, while a certain level of automation has been achieved in the presentation of roads, buildings and other cultural features for the production of digital map data, the automatic extraction of these features from images has not been achieved. The availability of high resolution digital satellite and multispectral aerial images, coupled with the community’s increasing need for more detailed, timely and lower cost spatial information for the production of digital maps and GIS (Geographic Information System) databases, has driven research on feature extraction over recent decades. While this research continues to develop new approaches to the extraction of features from images, no system has been so far demonstrated that enables extraction of features reliably under a range of image conditions and scales.