Part of the UK Ministry of Defence element of the STOW programme investigated the time and cost drivers pertaining to the entire process of the rapid generation of Synthetic Natural Environments (SNE) databases. Data requirements, products, information and systems were analysed to identify bottlenecks and gaps. Traditionally, construction of SNE databases is a time consuming and very labour intensive exercise. It involves a very high degree of effort to generate the required source terrain and feature data, and significant further effort to convert source data into a compiled SNE database.

Standard military datasets are typically used to provide the bulk of the data for a SNE database (e.g. DTED and DFAD). However, such datasets may not be available for the specific area of interest, they may be at an inappropriate scale, they require augmentation and they are likely to be based on out-of-date mapping sources. An alternative worldwide and up-to-date source is required. The new series of Earth Observing satellites are creating a large archive of up-to-date geospatial data. The major blockage has moved down the value-added chain and it is the conversion of data into information that has become the major time and cost driver.

An approach to automated feature extraction from EO imagery is presented which uses an object-orientated geodata model as the framework to store contextual knowledge and to use this in the control of feature extraction routines. The problem of geographic extraction has proved complex and ideally requires the incorporation of contextual clues similar to those used by human interpreters of imagery. Often the feature recognition algorithms work at local levels and in a bottom-up fashion and lack the higher level control that would allow a more global understanding of parts of the image. The paper proposes a control strategy that incorporates both the global and local views.

The geodata model comprises a class hierarchy representing the features under study and their likely relationships. Each class of object within this model contains criteria that need to be satisfied in order to strengthen the belief that an instance of that object type has been recognised. The criteria cannot be rigid and the system must be able to control partial recognition of objects and identify conflicts. The system described will apply these ideas to the problem of geographic object recognition, focusing on the specific requirements of linear feature extraction.