ComplexScan AUTOBOX v1
Hello! My name is Vladimir Matiasevich. I am an R&D engineer specializing in hardware, software, and AI solutions. Additionally, I have significant experience in 3D modeling and visualization.
Today, I will tell you the story of the creation of ComplexScan AUTOBOX v1 — a professional system for photogrammetric 3D scanning. This project became a logical continuation of the development of a series of transparent glass turntables for 360-degree photography, known as ComplexScan Transparent, and an Open Source DIY project called PhotoPizza.
In 2019, I participated in the launch of the objet.art project (ArtClub Digital Heritage). Our primary task was to develop a 3D scanning technology that would provide the highest accuracy and quality in visualizing art objects, as well as historical and cultural artifacts.
In addition to the high quality of the 3D models themselves, this technology needed to include the creation of high-resolution textures accurately reflecting the optical properties of materials and colors, which is critically important in this field. All of this is necessary for specialists in history and art to conduct in-depth analysis, develop restoration plans, preserve and utilize 3D copies, and determine optimal exhibition methods. The accuracy of the 3D model had to achieve a submillimeter level with an error range from 1 to 0.1 or 0.001 mm, depending on the size of the object. At the same time, there was a need for flexible configuration and scalability of all processes for applying the technology in different conditions.
We started by studying existing solutions. Essentially, we were choosing between photogrammetry and laser scanning. Laser scanning was not suitable for us as it required purchasing expensive scanners and proved to be less effective when working with objects of complex geometry or reflective surfaces. This led to the appearance of artifacts and inaccuracies in the model. Handheld laser scanners usually come equipped with low-quality cameras, negatively affecting the overall quality of textures.
Here is an example of the "ceiling" of the available laser scanning technology at that time:
As you can see, when zooming into such a model, both the polygonal mesh inaccuracies and the low quality of the raster texture are noticeable. Now let’s look at the photogrammetry as an alternative.
Photogrammetry is a technology for building a three-dimensional model based on a series of raster images of an object, viewed from different angles.
For further tests, I created a set of images of objects with various characteristics on the PhotoPizza turntable. Some of the most popular software tools that I tested included:
- RealityCapture
- Agisoft Metashape
- Meshroom
- 3DF Zephyr
- Autodesk ReCap
- Pix4D
Here’s an example of how the technology works:
And here is what the result looks like:
Ultimately, we chose the solution from RealityCapture because it provided the highest quality textures, great model detail, and had a high processing speed for raw data.
Photogrammetry allows scans to be performed at any resolution, as the quality of the obtained data depends solely on the camera resolution, the type of lens, and the total number of photos taken. For example, using photos taken by a microscope, one can create detailed 3D models of tiny objects. Furthermore, photogrammetry can work at astronomical scales: using images taken by telescopes to create 3D models of celestial bodies. This makes this method very flexible and versatile.
Example of RealityCapture in action:
But the main thing is, of course, obtaining the original data — high-quality photos taken from different angles and with optimal lighting conditions. These images serve as the foundation for accurate model building and texture creation.
So, let’s formalize the tasks of the technical process and describe the challenges:
- Lighting brightness issue. Objects need to be photographed with a partially closed aperture to achieve the necessary depth of focus. Additionally, it was often necessary to use a polarizing filters to exclude unwanted glare. Existing constant light sources on the market were not bright enough for such conditions and were generally cumbersome. LED panels, which were only available from China at the time, also lacked sufficient power.
- Complex object topology. To capture all the nuances of shape, the horizontal shooting angle needs to be dynamically adjusted. Thus, a solution with drive control, at a minimum, on two axes is required.
- Complex object textures. Glass, polished metal, bone, ceramics – all of these should look natural but possess very different optical properties. Transmitting material properties is one of the greatest challenges in 3D scanning.
- Accurate color reproduction. Data for color correction must be applied to the final texture in strict accordance with reference color keys.
- Shadows. Ideally, there should be no shadows on the texture of the 3D object at all. For realism, shadows are added later during rendering.
- Post-processing of 3D models. Texture retouching, retopology, sculpting, resizing and calibration.
- Data storage. Raw data of a single object can occupy up to 200 GB.
Considering the mentioned tasks, a decision was to develop new specialized equipment for automating 3D scanning using the photogrammetry method.
We applied a modular approach. Light sources were divided into active tiles, each of them could connect to another to increase the total area of the light matrix. This allowed the flexible scaling of the scene. Each light source included a wireless controller capable of changing the illumination scheme in real-time. Active cooling was implemented along with overheating sensors for automatic shutdown upon reaching a set temperature. Polarizing filters could be attached to the light sources for cross-polarization lighting. To fight unnecessary shadows that we don't want on the textures, even illumination from below was implemented.
Here’s a 3D model of the prototype:
Then we proceeded to manufacture the components and assemble them. It’s great to have your own laser cutting machine:
For our new modular architecture, we developed special modular software capable to manage any number of drives and lighting modules:
Next, we tested the working prototype on several objects. Here are the results:
Moving forward, we organized shoots at the State Museum Hermitage (St. Petersburg, Russia):
Before passing the equipment and refined technology for routine work, I needed to resolve a number of organization and process optimization issues. I've personally scanned several Japanese netsuke figures for the ArtClub and the Hermitage Museum:
By the way, talking about complex topology, notice how we processed those holes in netsuke figures.
The holes in netsuke figurines are called himotoshi. These holes were used to pass a cord through, which attached the netsuke to the kimono's obi along with small pouches or inro (small boxes).
After the project presentation, we moved forward. To scale the processes, we needed to build a team of 3D modelers and photographers. I've created the test assignments and reviewed them. We found the right people and continued to improve the technology together. We worked with very interesting objects of material culture, touching history with our own hands. Each new artifact presented us with a new challenge posed by complex materials and shapes.
For example, this dagger with a handle made of white jade: intricately carved surfaces are inlaid with precious stones: