Toward Sensor-Based Context Aware Systems

Yoshitaka Sakurai, Kouhei Takada, Marco Anisetti, Valerio Bellandi, Paolo Ceravolo, Ernesto Damiani, Setsuo TsurutaSensors 2012 · DOI

TL;DR

This paper introduces a novel methodology for interpreting sensor data and reconciling different types of uncertainty to build robust context-aware systems. By representing contexts as annotated business rules and using a flexible approach to handle fuzzy and probabilistic uncertainties, the authors demonstrate a practical system for real-time decision-making in healthcare scenarios.

What Problem Does This Paper Address?

Context-aware systems aim to adapt their behavior based on the user's environment and situation. A significant challenge in this field is accurately interpreting sensor data, which is often prone to uncertainty. This uncertainty isn't just about sensor accuracy; it arises from multiple possible interpretations of the same sensor reading, each with varying degrees of confidence. Traditional methods struggle to reconcile these diverse uncertainty models, leading to potential errors in decision-making. The paper tackles this by proposing a method to integrate sensor data with human knowledge (business rules) and manage the inherent uncertainties in a unified framework.

Methods

The core of the proposed approach lies in its methodology for sensor data interpretation and uncertainty reconciliation.

  • Context Representation: Contexts are defined using an extension of the Semantics of Business Vocabulary and Business Rules (SBVR). This allows for formal, unambiguous representation of domain entities, their properties, and constraints, which can be translated into First-Order Logic (FOL).
  • Sensor Data Interpretation: Sensor readings are interpreted to generate events. Each event is labeled with its appropriate type and, crucially, its level and type of uncertainty (e.g., fuzzy or probabilistic).
  • Uncertainty Reconciliation: The paper proposes a technique to reconcile different uncertainty types. Instead of directly combining uncertain assertions, uncertainty is moved from events to business rules. This is achieved by generating "combs" of standard Boolean predicates, effectively converting uncertain information into crisp conditions for rule evaluation. This avoids the complexities of directly conjoining assertions with heterogeneous uncertainty models.
  • Context-Based Reasoning: The system uses a Context-Based Reasoning (CxBR) engine. This engine processes the interpreted events and the formalized business rules to determine the current context and trigger appropriate actions. The system supports both Closed World (CW) and Open World (OW) assumptions for knowledge representation, allowing for flexibility in how missing information is treated.

Datasets

Not discussed in the paper. The case study uses simulated heartbeat sensor data.

Key Insights and Ideas

  • Uncertainty as a First-Class Citizen: The paper emphasizes that uncertainty in sensor interpretation is not just noise but a fundamental aspect that needs explicit modeling and reconciliation.
  • Bridging the Gap Between Sensors and Rules: By translating sensor events into annotated rules and then evaluating these against business rules, the system effectively bridges the gap between raw sensor data and actionable intelligence.
  • Moving Uncertainty to Rules: The technique of moving uncertainty from events to business rules via Boolean predicates is a clever way to simplify the reasoning process and avoid direct composition of disparate uncertainty models. This allows for a more robust and less error-prone decision-making process.
  • Hybrid CW/OW Assumptions: The ability to handle both Closed World and Open World assumptions within contexts adds significant flexibility, allowing designers to choose the most appropriate reasoning paradigm for different situations (e.g., critical vs. non-critical scenarios).
  • Efficiency through Crisp Predicates: By converting uncertain events into crisp Boolean predicates for rule evaluation, the system achieves efficient reasoning without the computational overhead of complex fuzzy or probabilistic logic engines at the rule execution level.

Key Results

The paper demonstrates the feasibility of its approach through a case study in a hospital setting, monitoring critical patients.

  • Successful Context Switching: The system successfully switches between "NORMAL" and "EMERGENCY" contexts based on heartbeat sensor data, controlling access to a radiology room.
  • Mimicking Fuzzy Controller Behavior: The proposed crisp predicate approach effectively mimics the behavior of a standard fuzzy controller for heartbeat classification, as shown in Table 1.
  • Significantly Lower Processing Time: The crisp evaluation approach achieved a much lower processing time (e.g., 0.000014 s for a crisp evaluation vs. 0.000813 s for fuzzy evaluation) compared to a benchmark fuzzy controller, indicating improved efficiency.
  • Robustness to Uncertainty: The methodology allows for handling different types of uncertainty (fuzzy and probabilistic) and reconciling them for consistent decision-making.

Limitations and Future Work

  • Manual Intervention: While the approach aims for automation, the initial selection of uncertainty models and the definition of crisp predicates (WCGE) still involve some degree of manual intervention by the modeler.
  • Ad Hoc Selection of Intervals: The paper notes that the intervals for the WCGE predicates might need careful selection, sometimes requiring an "ad hoc" approach, especially for fuzzy uncertainty.
  • Future Work: The authors plan to apply their approach to more complex domains like user behavior interpretation in computer-assisted collaboration and e-learning systems. They also aim to address the interpretation of user behavior and awareness.

Summary updated

BibTeX
@article{sakurai2012toward,
  author = {Yoshitaka Sakurai and Kouhei Takada and Marco Anisetti and Valerio Bellandi and Paolo Ceravolo and Ernesto Damiani and Setsuo Tsuruta},
  title = {{Toward Sensor-Based Context Aware Systems}},
  journal = {Sensors},
  year = {2012},
  doi = {10.3390/s120100632},
}
Figure