I am standing in the auditorium of an old abandoned German hospital. I can see dust across the room as it falls through a beam of light coming through a window. Above me, I can see the rafters of the abandoned architecture as if they were actually thirty feet above me. Scuff marks, upended tiles, and piles dirt across the floor indicate that the building has been uninhabited for several decades. An old abandoned piano is situated in the center of the room. What I see around me is unequivocally real. Well, as real as photographic reality capture can get. The program created by Realities.io hosts recreations of several different environments from across the globe.

realities_piano

Simply put, "photogrammetry" is the science of making measurements from photographs. Over the past few years, photogrammetry has become a popular method for recreating real objects, locations, and people as 3D models. Recognizing that the lines between real and virtually real blur more and more everyday, we sought out to improve our photogrammetry capabilities to allow us to take real elements into less-than-real environments. Realistically captured elements could be used in architecture to show complicated designs in virtual reality or to preserve past creations as you would with a photograph.


Before discussing how we went through this process, let's talk about drawbacks. The photo capture process can take a significant amount of time. Some photos may need to be eliminated, retouched, or entirely removed from the series of photos in order for the final photographic recreation to be realistic. Additionally, the 3D mesh that is constructed by the process can have significant errors, artifacts, and glitches. Some objects are easier to capture than others; reflective surfaces, people, and objects shot with moving backgrounds are notoriously difficult to recreate. That being said, the software used to stitch these objects together is rapidly improving.

Our goal was to use photogrammetry to recreate my self in VR as a 3d object so that I could be used as a stand-in character for realistic simulations. Rather than using mannequins, cut-out character silhouettes, or stylistic video game-y assets, we thought that using realistically captured 3D figures would have a greater impact.

Don't worry, it gets weirder.

Source: Tested.com

Source: Tested.com

In this article by Tested, the crew assembles a series of flat lights that provide clean, even, and balanced lighting on the model's face. For reality capture, you want your original 3D model to be lit as flat as possible. Lights, shadows, and reflections are all calculated in the VR simulation program (Unreal Engine, Unity, Cryengine, etc.)

Avoid reflections, bright lights, and moving elements. Any sort of movement in the scene can cause issues with the photo calculation. Basically, avoid everything that we did in our first attempt:

In our first attempt, Callum captured close to 200 pictures from varying angles around the room. The most difficult part of this attempt was getting me to keep my arms parallel to the floor for the duration of the shoot - a much more difficult feat than I had imagined. I did not initially understand how difficult it can be to remain perfectly still for several minutes. Optimally, the person being captured should not blink, talk, or turn their head. Breathing is acceptable though not encouraged. The model that was stitched together in Autodesk Remake had serious calculation issues caused by movement in our setting, so we gave it a second attempt. 


photogrammetry_pulse_design_group_polys

In our second attempt, we propped my hands with tripods to ensure that they would be perfectly parallel to the floor. After the capture, we simply removed the tripods from the 3D file. The 3D model had a significantly higher level of quality and resolution than our first attempt.

After exporting to an FBX file format to 3DS Max, we processed and rigged the 3D model for VR. The file is extremely high-poly (solid 3D models are composed of meshes of polygons. This had hundreds of thousands of polygons.) For optimization, we could manually reduce quality from parts of the image. In a scene with multiple photogrammetry captured characters, it would be essential to reduce the polygon resolution of as many models as possible. Additionally, I was able to process and touch-up the created photographic textures in Photoshop to ensure that lighting and shadows were even - especially across the front of the 3D model.

In the future, photogrammetry will become easier as the means for positional calculation become more advanced. If we were to have captured our images with higher resolution cameras and with uncompressed file formats, we would have had less problems with compression and color depth. Perhaps new phones with multiple camera sensors will bring photogrammetry to the masses with easy-to-use tools for development and distribution.

photogrammetry_scene