Reducing context uncertainty for robust pervasive workflows

Mobile computing devices equipped with sensors are ubiquitously available, today. These platforms provide readings of a multitude of different sensor modalities with fairly high accuracy. But the lack of associated application knowledge restrains the possibility to combine this sensor information to accurate high-level context information. This information is required to drive the execution of applications, without the need for obtrusive explicit human interaction. A modeled workflow as formal representation of a (business) process can provide structural information on the application. This is especially the case for processes that cover applications with rich human interaction. Processes in the health-care domain are characterized by coarsely predefined recurring procedures that are adapted flexibly by the personnel to suite specific situations and rich human interaction. In this setting, a workflow managment system that gives guidance and documents staff actions can lead to a higher quality of care, fewer mistakes, and a higher efficiency. However, most existing workflow managment systems enforce rigid inflexible workflows and rely on direct manual input. Both is inadequate for health-care processes. The solution could be activity recognition systems that use sensor data (e. g. from smart phones) to infer the current activities by the personnel and provide input to a workflow (e.g. informing it that a certain activity is finished now). However, state of the art activity recognition technologies have difficulties in providing reliable information. In this thesis we show that a workflow can aid as source of structural application knowledge for activity recognition and that the other way around, a workflow can be driven by context information in a way reducing the need for explicit interaction. We describe a comprehensive framework – FlowPal– tailored for flexible human-centric processes, that improves the reliability of activity recognition data. FlowPals set of mechanisms exploits the application knowledge encoded in workflows in two ways. ?Con (StarCon) increases the accuracy of high-level context events using information from an associated workflow. Fuzzy Event Assignment (FEvA) mitigates errors in sequences of recognized context. This way FlowPal enables unobtrusive robust workflows. We evaluate our work based on a real-world case study situated in the health-care domain and show that the robustness of unobtrusive health-care workflows can be