BUILDING RECONSTRUCTION CASE

Building Reconstruction Case

Building Reconstruction Case

Proposal for the Safe Removal of Existing Building

There is a growing interest in the continuous data collection of buildings, driven in part by recent advances in sensing technologies. Although commercial solutions are available for monitoring and preventative maintenance of mission-critical HVAC, lighting, security or circulation systems, these cover only a small fraction of the information that may be useful in assessing and improving building performance. What is still lacking is a systematic and comprehensive approach to collecting state information throughout the building life cycle. If overall facility performance is to be tracked and analyzed, it appears that 'inert' or 'grey' matter in buildings, such as building and space enclosures, furniture, manually operable windows and doors, should be included as well. Although generally thought of as static, these entities change considerably over time. For example, studies on churn rates and churn cost in office buildings suggest that significant physical changes may occur in workplace configurations over relatively short time periods. Through appropriate sensing infrastructures, entities affected by such changes could be made available for detailed performance analysis. This implies models of existing buildings that are updated frequently - ideally in real-time - and thus automatically. Manual model maintenance is a significant obstacle towards timelier, more comprehensive and accurate building models particularly in facility management and building automation applications.

Automated, sensor-driven reconstruction of building models may be organized as follows (Fig. 1). Traditional as-built architectural and engineering drawings provide the basis for a data repository of building objects that are customized and assembled on-site. This may include descriptions of layers of materials that make up walls, junctions, structural systems, and so on. Similarly, products manufactured and assembled off-site are stored in manufacturers' databases. Together with object data collected on-site, which reflect changes occurring during construction, operation, refurbishment, or demolition, these sources inform the building model reconstruction process. Whereas the two product databases are fairly static, continuous sensor readings trigger dynamic updates of a building model.

Fig. 1: Building model reconstruction process

Automated reconstruction and recognition of objects or scenes from sensor data has been researched extensively in computer vision and related fields. Most work deals with 'natural' scenes, that is, no or only minimal a priori knowledge about a target scene is assumed. However, progress with respect to potential applications in the building domain has been slow. On the other hand, the availability of smart sensing technologies - some of which employ computer vision - suggests the feasibility of scenes enhanced by markers or tags and tracked by sensors. Examples include electromagnetic, radio frequency, ultrasound and optical/vision-based technologies. Compared with natural scenes, potentially richer models could be derived from enhanced scenes because tags usually either incorporate or refer to class or instance information related to the objects to which they are attached. An object's identity or surface properties, which are relevant to physical asset management and lighting simulation, for example, are often difficult if not impossible to reliably extract from natural scenes. Therefore, the work presented in this paper adopts ...