4.1. Back to the Research Questions
Within the given framework of facility management, the results support that a 3D viewpoint that maximizes the visibility of 3D geometric objects inside the viewport improves the effectiveness (both in terms of success rate and accuracy) and certainty of a visual counting task compared to the traditional four side points of view. The statistical analyses even show that these two criteria are highly enhanced (p-value < 0.001). This aspect is essential from the perspective of including the 3D spatial data visualization in the decision-making procedures, especially when it comes to safety procedures. In the case of fire detection or propagation, the company in charge of the building safety might, for instance, provide a quicker and better evaluation to the emergency services. The deployment of required resources (both human and material) might then be enhanced and lead to facilitating the work in the field.
However, the hypothesis that the proposed point of view makes the visual counting of objects faster compared to a single side point of view could not be demonstrated at the 5% significance threshold. However, we can still notice that the 3D best point of view shows a lower variance, which may be explained by a higher consistency in the visibility of 3D geometric objects within the viewport. Moreover, the single side point of view greatly limited the choice of objects to be visualized due to their visibility requirement inside the viewport; the objects were usually close to each other, which may have facilitated the visual task achievement and reduced its completion time. In contrast, the precomputed point of view was applied on a more scattered spatial distribution of objects, which is more representative in practice. As a result, although no significant effect could be found, the proposed point of view still seems a promising solution for conducting a visual selective task as it still performs the task faster than the four traditional side points of view and is more suitable for actual scenarios in facility management.
Finally, the visual counting time per object linked to the best 3D viewpoint is not influenced by the background training, the decision-making level, and the experience in 3D visualization. The same applies to the certainty degree of the 3D best point of view, except for the strategic level for which the number of observations per category did not meet the statistical test conditions. To deeply analyze the effect of this decision-making level on the certainty degree, a higher number of participants is required and is vital to develop user-centered design strategies.
4.2. 3D Viewpoint in the 3D Geovisualization Process
Within the 3D geovisualization process, the results tend to support the key role of the 3D viewpoint in achieving specific visual tasks. While the visualization techniques applied to the 3D building model were kept constant throughout the experiment, we show that the achievement of a selective task (visual counting) can be greatly enhanced from a suitable 3D point of view. Initially included into the rendering aspect as variables of vision [34
], the camera settings should be removed and considered as an external processing stage as they clearly impact the completion of the visual task independently of the mapping and rendering aspects.
Beyond the theoretical outcomes, this work also proposes a first RESTful web application for managing the 3D viewpoint of spatial data. Developed as a client-server application that can leverage the power of remote computers, the application could become a promising operational solution to be incorporated to existing online 3D viewers. This architecture could even be deployed on any kind of device (from desktop computers and laptops to tablets and smartphones), as the processing stage is performed remotely. The only parameters sent by the server to the client are two sets of three dimension coordinates that automatically locate and orient the camera inside the 3D scene. Note that these two sets of coordinates could even be incorporated into the visualization of the OGC Web Terrain Service as a powerful way to improve the understanding of the 3D scene [35
]. For that purpose, the 3D coordinates of the best camera location should be converted into the OGC specifications: The distance from the point of interest (which is the 3D vector that fits the camera viewing direction provided by the algorithm), the pitch, the yaw, and the raw angles from this point of interest.
4.3. Limitations and Perspectives
First of all, only 36 experts participated in the survey, which is too small for the development of real user-centered design strategies. Consequently, the results must be interpreted with caution and a higher number of participants is required to achieve a greater confidence in the findings. Whilst the questionnaire was only deployed on the web for facilitating access during the interviews, no publicity actions were undertaken to introduce the questionnaire to a wider audience. Yet, this latter step should be required in the future in order to increase the sample size, although it does not guarantee the same degree of reliability as for interviews. Note that professional social networking sites (e.g., LinkedIn) could also be used so as to reach worldwide experts from the field.
Then, the visual counting task was only carried out on the windows of the 3D building, i.e., on objects visible from the outside. A more practical use case in facility management also requires 3D assets located inside the building, e.g., temperature or carbon monoxide sensors, ducts, and cabling. Since these objects are fundamentally occluded from an exterior 3D point of view, additional visualization techniques should be simultaneously applied (see [36
] for a taxonomy of 3D occlusion management techniques).
Then, the experiment was only performed on one specific selective task: Visual counting. However, the selectivity interpretation task includes other assignments, such as the location of one or multiple asset(s) and the evaluation of the spatial relationship between assets. The last one might require a step prior to the 3D viewpoint computation, for instance, the extraction of the 3D geometric object to be visualized (e.g., a 3D surface or volume representing the spatial intersection between two assets).
Furthermore, the maximization of the 3D geometric objects’ visibility within the viewport was used as the only indicator that defined the viewpoint optimality criterion. This choice was driven by the selective task (visual counting) and the visualization conditions (the objects of interest were clearly contrasted from their surrounding). In the future, new or additional indicators could be proposed to specifically meet the requirements of other visual tasks. For that matter, an existing list of descriptors can be found in [37
Beyond its use in facility management, the viewpoint management algorithm could be extended to the design phase of a building as a way to facilitate the understanding and solution of clash (or collision) detection. To date, current BIM software already provides algorithms to automatically detect clashes among architectural, structural, and MEP components [41
], along with a log and clash images. However, we noted that the software illustrations are not based on any specific design guidance. A 3D viewpoint management module linked to suitable visualization techniques could enhance the comprehension and solution of clash detections.
Finally, the present proposal is only applied on 3D spatial data visualized through 2D screens. With the ongoing and growing development of virtual reality, the automatic 3D viewpoint selection could assist immersive guided tours, such as fly and walk-troughs, for better interaction with the designed space. For instance, it could be a part of a BIM-game system, i.e., an approach integrating both building information modeling and gaming, to improve architectural visualization and education [42
]. In such systems, the algorithm could also enhance the visualization and understanding of simulations of physical building dynamics and behaviors of virtual building users.