SANY Sensors Anywhere

Data fusion and modelling is an effective way to add value to existing datasets of sensor measurements. This can be achieved via aggregation of datasets or inference of new data from relationships and patterns within existing datasets. Generic fusion provisions data fusion in a way that separates configuration and data from the algorithmic processing itself, allowing re-use of algorithms and pre/post-processing between datasets. Re-use of algorithms and techniques lowers the cost of development, configuration and deployment of fusion services.

Data fusion and modelling techniques are used to integrate observation data, contextual data and phenomenological models from different sources in order to obtain new environmental information where and when sensor measurements are not available. Observation sensors may include in-situ, airborne and space-borne types, while models may include deterministic and stochastic models.

In addition, data fusion numerical techniques provide a framework for integrating information uncertainties which are generated from sensor measurements and models with various inaccuracies.

In SANY we have developed four distinct types of reusable fusion services:

• Spatial fusion services : Kriging, Bayesian Maximum Entropy, socio-economic spatial correlation
  
• Causal fusion services : multi-linear regressions, neural networks
  
• Temporal fusion services : state-space modelling, Kalman filters
  
• Domain models : PECK predictive model of ground displacement during tunnelling activity

The following video demonstrations of SANY fusion services are available:

• Demonstration : Fusion WPS 1 of 2
  
• Demonstration : Fusion WPS 2 of 2

Innovation and impact:

• Use of OGC SWE metadata to allow on-demand 'Plug and play' sensor measurement datasets (automated pre-processing)
  
• Use of generic fusion to enable algorithm reuse (seperation of data and configuration from the algorithm itself)
  
• Propagation of sensor measurement uncertainty through fusion algorithms into final results (UncertML)
  
• Knowledge-assisted data fusion to improve accuracy and context of results
  
• Impact on OGC standards regarding use of processing services in sensor networks (SWE, UncertML)

These services have been implemented for environmental decision-supportapplications by various project partners in context with the SensorSA. They were then validated under multiple risk domain applications. These included pilot applications specialising in the prediction of microbial risk of exceedance in bathing waters in the Gulf of Gdansk (Poland), atmospheric pollution risks and false alarms in the City of Linz (Austria) and underground risks of subsidence in the City of Toulon (France).