www. infofusion.se Information Fusion Requirements on Databases Ronnie Johansson
infofusion Principles of data fusion automation Richard T. Antony (JDL DFG member) Artech House, 1995 470 pages It’s like a thesis on data fusion algorithms, problem-solving and database support There is reason to believe that the book is focused on target tracking type defense applications (spatial and hierarchical reasoning) Focusing on Ch 6 ”Database requirements” in this discussion.
infofusion Declarative Short-term: signals, sensor data, images Medium-term: clusters, tracks, situations Long-term: doctrine, soil type Procedural (long-term declarative knowledge w. control) Knowledge about how to reason: rules, pattern- based classification Declarative and Procedural makes up 16 classes of fusion algorithms (e.g., class I only relies on short-term knowledge, class VIII is general machine learning) Algorithm knowledge incorporation
infofusion Infrastructure consideration Higher-level fusion algorithms (i.e., relying on long-term and procedural knowledge) may be: Robust, Context-sensitive, and Efficient (in computational requirements) However: Requires more complex algorithms and may place heavy demands on DBMS
infofusion Dependence on infrastructure Ex: Problems with ordinary DBMS A road network stored as a vector of vertices Target tracking alg that depends on the distance between the target and the closest road – might require an exhaustive search of all vertices. This might be too slow for real-time tracking
infofusion Storing declarative and procedural knowledge Databases must support storage, maintenance and query of both types of knowledge. Declarative know. datastruct: tables, semantic networks, decision trees, lists, etc. Procedural know. datastruct: pattern/action rules
infofusion Claim 1.Lack of efficient database support for spatial, temporal and hierarchical reasoning is an obstacle to sophisticated fusion algorithms. 2.Linear indexing not sufficient for data search and manipulation.
infofusion Database models: Relational Pros: More general than older models (hierarchical and network models) Physical and logical data independence Standardized query interface Runs on numerous hardware platforms Cons: Table the only representation structure Joins can computationally expensive Spatial or combined spatial and temporal data may be inefficient to both search and manipulate. Table-based data model cannot preserve complex semantic relationships among data.
infofusion Geographical information systems (GIS) Pros: Supports storage and retrieval of spatially organized information. Supports spatial search and 2-D set operations. Cons: Does not support temporal reasoning
infofusion Object-orient databases ”While OODBs conceptually supports sophisticated [higher-order] problem solving approaches, current [OODBs] provides limited support for the maintentance, query or manipulation of spatial objects [and especially not for real-time applications].”
infofusion Algorithm requirements 1.Human problem-solving metaphor 2.Algorithmic issues
infofusion Human problem-solving metaphor ”Biological systems maintain a dynamic situation awareness wrt 3-D space by fusing sensory-derived information with a priori using multiple level of abstraction analysis.”
infofusion Algorithmic issues Spatial reasoning – entities of interest are often spatially distributed Hierarchical reasoning – abstract concepts, e.g., situations composed of simpler elements Temporal reasoning – states and situations typically change over time
infofusion Temporal reasoning Implicit – time-stamped data, filtering Explicit – causality of events
infofusion Algorithmic issues Support for retrieval from database that is dependent on both spatial and temporal aspects.
infofusion Conclusions ”… the effectiveness and efficiency of data fusion systems can be enhanced by the development of highly robust, context-sensitive and fusion algorithms that in turn are supported by database systems that both facilitate alg. development and enhance alg. efficency.”