Academia.eduAcademia.edu

Figure 1 – uploaded by Koen Van Balen

See full PDFdownloadDownload figure
Figure 1 Left: Section used to conduct work on a building dating from 1340; right: emergency recordinc

Figure 1 Left: Section used to conduct work on a building dating from 1340; right: emergency recordinc

Related Figures (4)

consists of streamlining the interaction with a user, who can indicate special regions of interest and can help the system to increase the efficiency, e.g. by manually eliminating trees in front of walls. For every selected image in a sequence, a depth map is calculated, Figure 3. A first rough visualisation helps the operator to get an impression of the results. The separate depth maps are first transferred to point clouds using the camera parameters. From this moment on some typical tools from Reverse Engineering point-cloud manipulation are used. First point clouds from different images in one sequence are merged together, Figure 4. On this level, decimation techniques can be used to reduce the model to a workable size. In an analogue way different sequences can be combined. Every step in the process starts with preset parameters that are typical for the filter. Operators can change them, implying that the result will depend on their needs and interests. It also allows them to construct a model that is in line with the performance of their computer. The result of this component is a meshed model, textured with the images that were used to compute it. Methods to extract the optimal texture from the images are under investigation. It is important to keep track of the accuracy of the results. The outcome of the automatic correspondence analysis already contains information about the accuracy. It is clear that every filter applied to these results must keep the results within the acceptable range. The general structure of this process should be cyclic rather than linear. The system must be able to go back and forth to adapt the results according to the requirements of the user.
consists of streamlining the interaction with a user, who can indicate special regions of interest and can help the system to increase the efficiency, e.g. by manually eliminating trees in front of walls. For every selected image in a sequence, a depth map is calculated, Figure 3. A first rough visualisation helps the operator to get an impression of the results. The separate depth maps are first transferred to point clouds using the camera parameters. From this moment on some typical tools from Reverse Engineering point-cloud manipulation are used. First point clouds from different images in one sequence are merged together, Figure 4. On this level, decimation techniques can be used to reduce the model to a workable size. In an analogue way different sequences can be combined. Every step in the process starts with preset parameters that are typical for the filter. Operators can change them, implying that the result will depend on their needs and interests. It also allows them to construct a model that is in line with the performance of their computer. The result of this component is a meshed model, textured with the images that were used to compute it. Methods to extract the optimal texture from the images are under investigation. It is important to keep track of the accuracy of the results. The outcome of the automatic correspondence analysis already contains information about the accuracy. It is clear that every filter applied to these results must keep the results within the acceptable range. The general structure of this process should be cyclic rather than linear. The system must be able to go back and forth to adapt the results according to the requirements of the user.
Figure 6 Annotation using themes and layers Selection criteria on an un-annotated geometrical model can use only geometrical constraints, e.g. enclosing polygons. When more information is added, selection criteria can be more elaborate: e.g. select all entities within this polygon that belong to layer X in theme Y. Furthermore, Boolean operations on previous selections can be made. Specific geometrical querying is implemented to measure lengths and areas on the model, including an estimate for the absolute error of the result. A very special type of querying the database is the visualisation. As not all information from all themes can be seen at once, a view on the database can present only a partial view. In a first phase, one theme and numerous vector themes can be visible. The texture colour is changed according to the layer colour of the triangle in this theme. The visible entities and layers in a window can be exported to common formats as DXF and VRML. This widens the possibilities for distribution to the public, as standard browsers can view them. These facilities permit to derive all types of interrelated information from the model.
Figure 6 Annotation using themes and layers Selection criteria on an un-annotated geometrical model can use only geometrical constraints, e.g. enclosing polygons. When more information is added, selection criteria can be more elaborate: e.g. select all entities within this polygon that belong to layer X in theme Y. Furthermore, Boolean operations on previous selections can be made. Specific geometrical querying is implemented to measure lengths and areas on the model, including an estimate for the absolute error of the result. A very special type of querying the database is the visualisation. As not all information from all themes can be seen at once, a view on the database can present only a partial view. In a first phase, one theme and numerous vector themes can be visible. The texture colour is changed according to the layer colour of the triangle in this theme. The visible entities and layers in a window can be exported to common formats as DXF and VRML. This widens the possibilities for distribution to the public, as standard browsers can view them. These facilities permit to derive all types of interrelated information from the model.
Figure 5 Geometric manipulation
Figure 5 Geometric manipulation
Related topics:
Computer VisionPoint CloudMultimedia Data
580 California St., Suite 400
San Francisco, CA, 94104
© 2025 Academia. All rights reserved