dc.description.abstract
The quality and cost effectiveness of services in the building industry possess high potential for improvement. One significant approach in bringing out this potential is to conceive buildings as sentient entities that continuously adapt to changes in the environment.<br />A sentient building possesses a multi-faceted internal representation of its own context, structure, components, systems, and processes. This representation enables the self-regulatory determination of such a building's indoor-environmental status in accordance with the needs of its occupants. However, towards the realization of the sentient buildings, already acquired scientific foundations (theories, methods, and tools) must be transformed into a technically mature and industrially promising level. Specifically, such transformation must occur in three critical areas.<br />Firstly, the representational core of sentient buildings must integrate rather static building component class hierarchies (product models) with process-oriented systems controller hierarchies (process models).<br />Secondly, to achieve real-time building operation support and to avoid bottleneck situations resulting from manual model input and updating activities, the underlying product-process model must possess the capability to autonomously update itself. Finally, given the specific features and challenges of the building systems control domain (e.g.<br />multiple domains/systems, multiple levels of spatial hierarchy, contingencies of outdoor climate and occupancy behavior), proper control semantics (methods, rules, algorithms) must allow for scalable implementation schemes.<br />To provide a proof of concept for the feasibility of this transformation towards the realization of sentient buildings, a lighting control system is developed within the scope of a FWF project. The aim of the project is, concisely, to provide and maintain the most desirable lighting conditions in an office space. The research described in this thesis focuses on the second challenging item of the transformation within the project context. The second item implies that the associated representations must be self-updating, if they are to be applied effectively in the course of building operation and maintenance activities. This requires capabilities in the areas of contextual and indoor-environmental monitoring. In the lighting control system, the required monitoring capabilities arise in three major fields. Objects in the space must be identified, their locations must be sensed and occupancy information must be obtained. In addition to these monitoring activities, the prospective solution must comply with the building-specific requirements, where low-cost, low-maintenance, and scalability are crucial. In this dissertation, the study towards the realization of these capabilities is described. Prior to the implementation of a solution, available technologies are reviewed. With respect to the requirements, vision-based approaches were found to be preferable in terms of being software supported and system customizable. In our efforts for realizing such a solution, a Vision-based Object Location and Occupancy Sensing system (VIOLAS) is developed. VIOLAS extracts context information from the environment using image processing methods applied to the scenes captured by the cameras. VIOLAS utilizes network cameras for this purpose. These new technology cameras are feasible for buildings. They make use of the existing network installation without requiring additional infrastructure. They act like regular network devices, and convey camera images with standard Internet protocols. Through the same communication channel, they also enable the control of third-party devices like pan-tilt units that effectively increase the monitoring ranges. In addition to its primary objectives, the software implementation of VIOLAS must fulfill the aforementioned building-specific requirements.<br />Towards this end, the research proposes a common model that integrates hardware and software whereby the components are tied together via Internet. Network cameras constitute the hardware part of the system, and fit in this structure by conveying video images like as distributed network devices on Internet. Image Processing Units (IPUs) form the distributed software components. They are the programs that perform vision-based sensing and extract the context information by applying optimized image-processing and computer-vision methods on the images captured from the cameras. IPUs, implemented on different computers scattered across the facility, convey the context information to a central Application Server, where the parallel incoming results are combined, displayed to the operator, and concurrently conveyed to the lighting control system. In addition to enabling scalability and incremental growth, the distributed structure of the model enhances performance resulting from the parallel operations.<br />Additional function of the Application Server is to control the status of the components and dynamically assign active network cameras to active IPUs in such a manner that the workload is constantly balanced within the system. This arrangement provides a kind of self-organizing capability, and minimizes operator overhead. The resulting flexible and adaptive structure is highly suited to the requirements of control applications for sentient buildings.<br />
en