The project is organised into four main activities each of which makes contributions towards the realization of the proposed system Architecture:
These analyses investigate sketch maps as they are used in human-human interaction and analyze typical inaccuracies and errors in the drawings. The findings in related studies on human spatial cognition and in our empirical analyses lay the foundation for our following investigations. In general,we will, first, elaborate typically incorrect or inaccurate information in a sketch map; i.e. we describe systematic errors in spatial judgment which are a natural consequence of schematization and distortion in normal perceptual and cognitive processes. We do not focus on errors from drawing by hand. At the same time we will characterise the level of abstraction at which information can be considered as correct, i.e. can be equally found in a sketch and a metric map. For example, while the exact geometry of streets is distorted, the basic street network is usually correct. We also define what correctness means in this context, classify different kinds of errors or inaccuracies at different levels of granularity, and estimate their influence on sketch mapping.
Object recognition in sketch maps
One of the goals of SketchMapia is to be able to automatically extract geospatial information from sketch maps. Sketch maps are drawn free-hand on a piece of paper and captured into a digital representation, for example, using a mobile phone. The recognition and extraction of sketched features as well as the analysis of drawing inaccuracies such as overshoots and undershoots are done by computer vision methods. During object recognition, different features of free-hand sketch maps will be detected and classified semantically using the following approaches:
- object recognition
- text recognition
- domain knowledge modelling
Qualitative Description of Sketch Maps
Human knowledge about space is typically qualitative (Freksa 1991): Instead of absolute locations and their exact geometries, we recognize only few but significant relationships of spatial objects at a level of granularity which is important to solve tasks in our everyday life. Humans draw sketch maps from their observations and from their memories. Therefore, the set of relationships and attributes used to describe a spatial situation in a sketch map differs substantially from metric maps, which are based on exact and complete measurements. In this activity, we aim at developing a formal qualitative language to describe only cognitively relevant attributes and relationships in a sketch map at a sufficiently abstract level, such that distortion and schematization in the human cognitive process do not have a negative impact on the map alignment. With respect to the use case, we concentrate on two targets:
- To describe attributes and relations of spatial objects at the necessary level of abstraction at which the same information can be found in all sketch maps. This is the foundation for the integration of qualitative data in the tasks that follow.
- To describe attributes and relations of objects in a sketch map at the necessary level of abstraction at which the information is accurate enough to be aligned to a metric map. This is the foundation for the integration of quantitative and qualitative data in the tasks that follow.
Integration of Qualitative and Quantitative Data
Here we develop an analogy-based approach for bringing together qualitative data from sketch maps and quantitative data from metric maps. SketchMapia aims at the integration of data collected by the people, i.e. the integration of qualitative data available as sketch maps. Furthermore, sketch maps may be used to query the system. Before the qualitative sketch map query can be processed on the quantitative representation of a metric data, both representations have to be transformed in a compatible format, i.e. the quantitative data must be re-represented in a qualitative way. When both data sources are represented in the same qualitative format, analogical mapping is used to compare the maps and detect shared spatial objects and shared structures in both maps. This work package develops an approach to detect analogous elements between two or more sketch maps and analogous elements between a sketch map and a metric street map.The work here involves analogical mapping of qualitative data using appropriate analogy engines, integration of qualitative data with other qualitative data, mapping of qualitative onto quantitative data, and heuristics for data retrieval and scaling based on qualitative reasoning methods.