RSRU – Automated feature extraction in urban areas

Informal settlements are common physical entities within the makeup of South African cities. These settlements have increased in number from 1,049 million dwellings (in 1994) to 1,376 million (in 2004), and are projected to continue increasing to some 2,4 million by 2008. The existence of informal settlements is fast becoming a serious problem since they accommodate a large proportion of the urban population who live in sub-standard living conditions. In addition, the increasing trend in migration to urban areas inevitably leads to a shortage of basic engineering services such as water, sewerage and solid waste removal. Rapid urbanisation also places increased pressure on essential services such as health and education.

Informal settlements are known as shacks, squatter areas shanty towns and irregular settlements. Regardless of the name, common features that distinguish them from formal settlements are that they do not adhere to local building codes, they have either low or no levels of infrastructure, they are either poorly serviced or not serviced at all, they have no security of tenure and they are characterised by a rather non-functional layout.

One of the fundamental difficulties that authorities face when planning a response to the formation and growth of informal settlements is the lack of spatial and temporal data. Such data allow us to identify and quantify services and infrastructure, which are required to improve our understanding of settlement morphology, population distribution and emerging settlement patterns. Several reasons exist for the scarcity of data on informal settlements. Their dysfunctional structure and high building density make it hard to conduct surveys. The settlements are dynamic, with frequent population fluctuations – to the extent that the erection or removal of structures often happens overnight. Shifts in municipal boundaries and overlapping administrative responsibilities contribute to the confusion.

Additional challenges include deficits in manpower, funding and equipment. All these factors contribute to the difficulty of obtaining the data required for effective planning in and around informal settlements. To overcome some of these difficulties, the CSIR is investigating the use of satellite images to fill the gaps in the spatial data. The goal is to use QuickBird imagery (with a spatial resolution of 0,6 m) as a primary data source to map out the extent of informal settlements, simultaneously determining the specific settlement type in each area.

The first step was to use the QuickBird imagery to delineate homogeneous areas manually (see Figure X), from which urban settlement attributes were identified. These attributes relate to the settlement layout, housing structure, presence of engineering services and infrastructure existing within a particular settlement. The results of the manual extraction of the attributes showed that these could be identified from the QuickBird images, whilst establishing whether the settlements had access to different types of engineering services were inconclusive. Based on the identified attributes, a classification system of five classes was proposed to describe urban settlements. The classes include informal housing (IH), upgraded informal housing (UIH), formal suburb (FS), formal suburb with backyard shacks (FSB), and non-urban (NU).

Figure X

The next step was to develop an automated settlement classification procedure, in collaboration with the Meraka Institute of the CSIR. An intuitive approach to identifying settlement type would be to look at the distribution of building shape and size over a small region, say, 100 m by 100 m. Building delineation methods could potentially be used to extract this information automatically from satellite imagery, but the small structure size and variability of construction materials used in the informal settlements make this approach unreliable. Instead of trying to identify individual structures, a set of texture features can be used to describe the appearance of the settlements.

Texture features describe the pattern of adjacent light and dark regions of an image – these features can be likened to the perceived ‘roughness’ of a surface. An informal settlement with many small, closely-spaced structures will appear to have a finer texture than a suburban area with larger buildings. These features can be extracted automatically from the satellite images, after which a classifier can be used to identify what settlement type we are looking at. Researchers have developed a prototype of such an automated settlement mapping system – Figure Y illustrates the output produced by the prototype. Current research is focused on the development of automated feature extraction algorithms that describe settlements more effectively.

Figure Y

The ultimate goal is to build a system that can identify changes in settlement patterns automatically over time using satellite images, thereby alerting officials of potentially significant changes on the ground.