Quadcopter Photogrammetry

Nearing the end of my graduate program in media arts and computer science, I found myself stuck working on a thesis I no longer had much interest in. I had lost my motivation and feared I would end up in grad limbo with a project I couldn’t bring myself to complete and expectations, including my own, unmet. About this time a friend of mine had signed up for a class that was going to Cuba. This was an opportunity I couldn’t miss, so I signed up. This adventure led me to many others, including the genesis of what would become my new thesis.

I’ve long been an R/C flying enthusiast. In my boyhood, my dad and I built a small, gas-powered balsa wood plane. It was tethered to a string, and you could only fly in a circle. The poor plane didn’t survive its maiden flight. That concluded my R/C experience for many years, as we couldn’t afford to rebuild it. The price of all things R/C at that time made it cost-prohibitive for many.

A couple of years ago I discovered the hobby anew. I purchased a little R/C helicopter for my brother and was surprised by the quality, flight time, and maneuverability. I started doing some research and found a whole new, more affordable world of R/C. This revolution was mainly due to the advent and proliferation of lithium polymer batteries and brushless motors, which replaced expensive, messy gas motors and made electric models a more realistic proposition. And so my obsession began.

It started with small helicopters. Then larger helicopters built from parts. This led to airplanes, which was how I began doing FPV (first person view) flying. Soon after came tricopters and quadcopters, which provided full three-dimensional freedom of movement and a very stable platform for cameras.

Then came Cuba, photogrammetry, and liberation from my uninspired thesis. Photogrammetry is a method for creating 3D models of objects by taking a series (usually hundreds) of photographs. The concept is as old as modern photography. What has changed is the use of digital photos and software. The software takes all of the photographs and compares them to find matching points. Then the software uses these points to calculate depth.

Through my graduate program in media arts at New Mexico Highlands University, I traveled to Cuba for photography, photogrammetry, and an exchange of ideas. One goal was to make contact with the Office of the Historian, which is responsible for restoring the buildings of Old Havana.

Highlands University had been working with the Georgia O’Keeffe Museum for about a year developing the use of photogrammetry as a tool for conservation and preservation. We hoped to share these simple and inexpensive techniques with members of the Office of the Historian. We contacted them, learned more about what they do, and demonstrated the methods we had developed for documenting historic objects and sites using photogrammetry. They were very excited and offered us the opportunity to create photogrammetric models of several buildings and structures.

This was my first real opportunity to use photogrammetry in the field, and I too was impressed with what was possible. However, while working on documenting several structures, it became apparent that we were limited by taking photos at ground level, which created gaps in the images. Once we had rendered a preliminary model of Hotel Santa Isabel, I found that anything above the field of view would inevitably show up in the data as black holes rather than a solid 3D model.

I started thinking of different ways to get a complete view of the building. One obvious method would be to rent a hydraulic lift, but that could be costly and impractical in tight spaces. Helicopters might work but would also be cost-prohibitive. Then it hit me: I could use multirotor R/C aircraft to photograph the inaccessible areas. My passion for photography and the R/C world came together in a beautiful way.

When I returned from Cuba I immediately got to work. I had only recently started experimenting with building quadcopters, and the one I owned had seen better days. But I went ahead and started modifying it to carry a camera for my proof-of-concept build. My initial test used a GoPro Hero HD set for time-lapse and my house as the subject. I shot 200 pictures, and the results, though not beautiful, were very encouraging. I set out to create a purpose-built quadcopter as a stabilized camera platform to create photogrammetric models of large-scale structures.

From the outset, I was determined it would be affordable and accessible, and I hoped my idea could inspire and educate others. In its simplest form photogrammetry can be done with a compact digital camera and a laptop with surprisingly good results. But as the desired quality of the finished model goes up, the hardware requirements and processing time rise dramatically.

Once I had built the new quadcopter, I began testing and collecting data. It worked flawlessly. I collected hundreds of photos to be processed and turned into a complete 3D model of the historic multistory building that was my subject. When the photos had been processed and a complete model had been created in software, I concluded that my methodology was sound and completely viable as a useful tool for photogrammetry of large-scale structures.

In association with the field-testing, I created a blog to help anyone who might be interested in getting started with their own quadcopter. Uav3-d.info has articles on just about every concept of quadcopter flight so that this technology can be accessed by even the most uninitiated.

MY SETUP

MULTIROTOR: Custom built using parts largely from rctimer.com with an APM 2.5 controller board from diydrones.com. It has 30-amp SimonK ESCs (electronic speed controls) and 900kV motors with 10×4.7″ carbon-fiber props.

CAMERA: Canon PowerShot running CHDK custom firmware, which lets the camera shoot in RAW format and take photos automatically.

Photogrammetry Tips

1. A digital camera with fixed focal length is best.
2. Make sure your photos overlap 60% to 80%.
3. Take the photos horizontal to your object and at a uniform distance.
4. Uniform lighting is important for creating good models.
5. Process the images in Agisoft PhotoScan.
6. Low-quality models can be generated on a laptop. High-quality models require multi-GPU systems with 128GB of RAM.
7. For small models you can get away with 30 to 60 photos. Larger models (like buildings) may require several thousand images. The more photos, the better detail you can achieve.

This article appeared in MAKE Volume 37, page 42.