Log in / Register
Home arrow Computer Science arrow Augmented and Virtual Reality

D Catch

Autodesk 123D Catch provides a simple and user-friendly way to record threedimensional data without the need for specialized equipment or technical skills. It is accessible to anybody with a consumer camera. Autodesk 123D Catch finds common features on all the photographs and uses them to reconstruct a threedimensional scene. To create a model of the statue of Juulius, the eternal student (Fig. 4) and the mascot of our university, we took about 50 photographs sequentially around the statue, each showing the whole statue, uploaded them to 123D Catch and received a three-dimensional scene representing the statue. Such a three-dimensional scene can easily be exported as a mesh object and opened in Blender where it can be modified, and represented interactively to a user using Blender game engine.

As mentioned before, the process is straightforward, the reconstruction in 123D Catch requires no adjustments by the user. In our case, it took about half an our to upload the pictures and obtain the model, the result is very detailed and presented in Fig. 5 for Juulius and in Fig. 3 for the meerkat skulpture for comparison with the Kinect scan in Fig. 2.

Fig. 4. Photo of Juulius, the eternal student, mascot of Tallinn University of Technology

Python Photogrammetry ToolBox

The Python Photogrammetry ToolBox (PPT) is a GUI for several tools to simplify the reconstruction process. Using PPT only some mouse-clicks are necessary to obtain structure from the images, however the process is not as automatized as with 123D Catch. The components invoked by the GUI are Bundler and PMVS, CMVS. After MeshLab needs to be invoked to create the mesh. PPT can take a mix of pictures from different cameras, but for all cameras the width of the CCD sensor needs to be known.

The following steps are performed:

– check camera database and enter parameters if necessary

– run Bundler, in this case we used the siftvlfeat for feature matching and the images have been scaled down to 1200 pixels

– then PVMS was used to create the point cloud

– in MeshLab the Poisson Surface reconstruction is used, here it may be necessary to remove points belonging to the surrounding

The result of the reconstruction is shown in Fig. 6. The result is slightly less precise as the one obtained by 123D Catch, but this might be due to a nonoptimal choice of parameters or the scaling of the pictures.

Fig. 5. Mesh of Juulius obtained with 123D Catch


Creating three-dimensional reconstructions using VisualSFMs graphical user interface is similar to using 123D Catch. First, the user needs to select the photographs to work with, then, with two button clicks, it is possible to receive a sparse reconstruction based on those. With one more button click it is possible to receive a dense reconstruction, computing which takes time -? for us it took about two hours to compute a dense reconstruction of the statue of Juulius. It is possible to run VisualSFM on the command line and change some of its parameters, giving the user more control over the steps of the reconstruction process. The resulting point cloud with the camera positions is shown in Fig. 7. Being able to see the camera positions gives the possibility to see possible reasons in case the model has holes or lacks details in some areas, due to sparse photo coverage.

Found a mistake? Please highlight the word and press Shift + Enter  
< Prev   CONTENTS   Next >
Business & Finance
Computer Science
Language & Literature
Political science