There are many use cases for depth sensors and 3D scanning is actually one of the more niche ones. And the available software usually supports a select amount of sensors. Here are some of the affordable sensors supported by RecFusion.
The last two also support multi-sensor scanning:. With the latter you can capture from more than one sensor simultaneously and calibrate their positions in a fixed rig. This way you can scan objects and people without moving the sensors. For example, you can mount 3 sensors on a vertical rail and use a turntable for scanning people. Image Courtesy of RecFusion source. You can save these settings as a so called Workspace, which is handy if you work with different setups for different purposes.
The Volume Resolution is set in voxels. Which is quite unique but easy to understand once you notice the voxel size. In the sensor panel you can select any combination of color and depth resolution your sensor supports, although not al combinations work. I did notice that I sometimes got combinations to work when switching to a different depth or color resolution and then back to the one I originally wanted.
Manual exposure is only available on certain sensors, though. Finally, the depth cutoff setting can be helpful for scanning objects with a sensor that has a large scan range. If you want to use the sensor hand-held, which can be handy with or without a turntable to capture surfaces in hard-to-reach places, RecFusion offers both a start delay with countdown and fixed scan time setting. With this feature you can more easily walk around with a laptop or tablet in one hand and a sensor in the other.
As you can see the 3-way split screen shots the RGB feed, the depth feed and a 3D voxel point cloud of the combined results. All three work in vertical orientation but unfortunately the 3D view loses its orientation setting while scanning. Then it shows a real-time fusion preview in horizontal view. You can switch off the color view while scanning which can make is easier to track your progress. When doing that, RecFusion will try to recognize the camera angle and continue from there.
This worked as expected most of the times I tried it. Here the captured data is directly fused into a 3D model. This method relies on the graphics card GPU of the computer. Instead of real-time fusion you can also opt to simply let the software record the image sequences from the depth and color camera.
You can then save and open this sequence file in RecFusion for offline, or non-real-time, reconstruction. According to this tutorial page on the application website, there should be an option to perform this offline reconstruction on the computers main processor CPU instead of the graphics processor GPU. The developer has let me know that the page is outdated and that CPU-reconstruction is now longer supported.
Unfortunately the native color capture is only done with per-vertex coloring, meaning that the color quality is strictly tied to the geometric volume resolution. This differs from programs that can project separately captured color stills onto the geometry using a technique called UV texture mapping.
You can view the finished scan with or without color applied, or view the wireframe. Most well known editing features are present in RecFusion, like cropping the model pictured below or selecting orphan elements automatically for clean-up.
If you do want to make 3D prints, RecFusion can also generate a pedestal called socket in the software. There appears to be hollowing feature indeed, but its located in the Export tab, which brings us to…. Saving your Workspace is strictly for settings. The model has the be watertight for this to work.
Of course, quality depends on the sensor you use. The per-vertex coloring method has its limits. But if you hit 3 on your keyboard you can switch the Sketchfab viewer to MatCap mode to see just the geometry. There are some imperfections at the loop-point on the left of the back shoulder where the new geometry is supposed to fuse with the starting point of the scan. But that was not an issue I consistently experienced. Overall the geometry quality is a typical depth-sensor scan.
As you can see that geometry is a lot smoother than the hand-held model, so adding some human motion can influence the scan result. As you can see the geometry is similar to the last RecFusion embed but the color quality is a lot more photo-realistic due to the mapped image files. Convenience classes for accessing OpenNI2 based sensors are included. There's also a sample showing how to use the reconstruction with RealSense D series sensors.
In addition, methods for calibrating multi sensor setups are available as well as basic mesh post-processing tools closing holes, hollowing out mesh, cleaning and smoothing are also included. It is also possible to directly access the mesh information in your program.
In the non-commercial free evaluation version the mesh export is disabled. View online documentation. The license includes one year of free updates and support. Please inquire for prices for purchasing multiple licenses at This email address is being protected from spambots. You need JavaScript enabled to view it. For questions and feedback you can use the RecFusion forum.
Old versions. Our website uses cookies to make your browsing experience better.
0コメント