Tag Archives: Drone

Model of my Deck

I entered the following in the WebODM (Open Drone Map) forum. Some of it is a bit technical for this blog, but thought it might be interesting to some people.

Just for fun, and to learn more about WebODM and Blender, I flew my DJI Mini 2 drone around my deck to create a model. My deck has trees on 3 sides of it and overhanging it. Flying between the tree branches to get some of the shots was a bit challenging. There was a bit of a pucker factor a few times when flying inches below a branch and the drone started drifting upwards! (The Mini 2 has only forward and downward sensors, which is good here – I could never have flown that close to the trees if there were active sensors the other directions.)

I shot 142 images and processed those and saw some areas that didn’t seem to have adequate coverage. So I shot another 49 images to fill in some areas. That improved the places I concentrated on, but it seemed that some other areas decreased in quality. The glass railing and adjacent sunroom windows and doors caused some oddities, as expected. One thing I found odd is that the deck and other items seem to be “reflected” in the undersides of the tree branches.

My processing system is Windows 10 Pro on a laptop with 64GB of RAM. I initially processed this using Docker/WebODM, but ran out of memory when I increased pc-quality to ultra. I then processed it in Windows native WebODM, and it processed in 24+ hours. The WebODM timer showed 36 minutes, so I don’t have accurate information…

I postprocessed this with Blender to clean up some of the extraneous parts of the model, but purposefully left most of the trees. To get the upload files under the 100MB limit for the free Sketchfab account, I decimated the model in Blender to 70% and converted the PNG files to JPG.

Processing parameters:

debug: true, mesh-octree-depth: 13, mesh-size: 1000000, min-num-features: 30000, pc-quality: ultra, resize-to: -1, verbose: true

Click on the image below to activate the model. Use your mouse buttons to change the view, and your scroll wheel to zoom in and out. Type “F”, or click the double-ended arrow in the lower right, to open it in full screen. (I highly recommend viewing it in full screen.)

PHOTOGRAMMETRY: 3D Models from Photos

(From Autodesk’s website:) What is photogrammetry?

Photogrammetry is the art and science of extracting 3D information from photographs. The process involves taking overlapping photographs of an object, structure, or space, and converting them into 2D or 3D digital models.

Photogrammetry is often used by surveyors, architects, engineers, and contractors to create topographic maps, meshes, point clouds, or drawings based on the real-world.

I’ve written in past posts about 360° panorama photos (360° Panoramas!, More 360° Panoramas!, and 360° Panoramas (again)). In a 360° panorama, the camera (the viewer) is at a single location looking out on the world. Today, we will visit what seems to be the opposite situation.

3D models are created by taking a series of photos of an object from many different directions. The object could be something small, like a sculpture. Or something large, like a movie set. Or something in between, like a building. The camera could be mounted on a tripod and the small model turned to different positions, or the camera could be moved around the small model to take many different views. For an even larger model, the camera could be carried by a drone, for instance, and moved around a very large area to take many images.

I’ve played with 3D models a bit over the last few years. Once you have acquired images of your target, they must be processed in some way to create a 3D object, usually a “mesh” of many triangles that simulate the original model. Much of the software to do this is relatively expensive (hundreds or thousands of dollars), or rented by the month. However, not all software is expensive. After looking at other options, I found Open Drone Map (or ODM). The original purpose of ODM apparently was to create maps and/or models from photos taken from a drone. However, the software doesn’t really care whether the camera was on a drone, or handheld, or on a tripod.

Using ODM, I was able to successfully process several sets of photographs I have accumulated over the last few years. My smallest models were created from about 40 photos shot with my cell phone and the largest I’ve created so far used a couple hundred photos shot with a drone. People successfully use ODM with 5,000+ photos, although that may take days to process, even on a powerful computer.

Once you have created a 3D model you must use special software to view it. Surprisingly, current versions of Windows do come with a simple 3D viewer, but it doesn’t seem to be very robust. There are also websites where the 3D model can be uploaded, then you can view the model with a web browser.

Below is one of the first models I created. It is a tabletop scene of a small wood manger. This model was created from 48 photos shot with my DSLR as I walked around the table, taking photos at different heights to be sure everything was visible. Click the “play” button, wait for it to load, then use your mouse left button spin the model around on your screen, and your mouse scroll wheel to zoom in and out. To see the model full screen, press the “f” key. (I recommend trying that – press the “f” key again to exit full screen mode.)

The photo below is one of the 48 photos that make up the model above.

Another 3D model I created is an interesting rock at the entrance to the Arizona-Sonora Desert Museum (ASDM). This one is created from 40 photos I shot with my cell phone as I walked around it several times.

I used several other programs to generate all of the models shown here. First is WSL – Windows Subsystem for Linux. The version of ODM I used runs on Linux, so this allowed me to run it in a Linux environment on my Windows computer. I used Blender to clean up (remove) the extraneous parts of the 3D images, which were then uploaded to Sketchfab. Other programs played more minor roles. Expect to see more about Blender in this blog in the future.

Night Drone Photography

With recent changes to the FAA rules, it is now possible for drone pilots to fly at night without jumping through as many hoops as before. To be eligible to fly at night, I had to take the update course for my “Part 107” certificate. I also must have an anti-collision light on my drone that is visible for 3 miles in any direction (that sucker is bright!).

When flying at night, one needs to be very aware of their surroundings so as not to hit something that you can’t see. It’s best to check out the location in daylight hours to be sure there are no wires or other such items that you might encounter.

I flew my first night flight a couple weeks ago just to try it out. I flew from my deck, which is surrounded (and partially covered) by trees. I know where they are, and where on my deck I am clear of overhead obstructions. Landing is the tricky part — making sure that I am not descending into the trees or onto my roof. I was successful in flying a short flight and taking a few photos.

Several days later I flew from the Edmonds waterfront. I walked up the beach until I was clear of other beach-goers and had a good place to take off and land (a flat, almost-level, rock). My goal was to get some shots showing the Edmonds Ferry at or near the dock and the city lights of Edmonds. I was successful.

I flew my DJI Mini 2, which is very light-weight and has a 12 megapixel camera. I would like to try again with my DJI Phantom 4 Pro, which is heavier and has a better quality 20 megapixel camera. I think the heavier drone will probably be a bit more stable, which will improve the sharpness of the photos taken with the slow shutter speed required. Although, looking at the photos, the sharpness is quite good considering the camera is “sitting” on a platform floating in the air, subject to wind and motor/propeller vibration. Shutter speeds were between one third and one second with ISO varying from 1600 to 3200. With the small sensor on the DJI Mini 2, these high ISOs made for somewhat grainy photos.

The photo below was shot as a series of nine RAW photos. The drone was positioned at one point in the sky, then three photos were shot using exposure bracketing (each photo with a different exposure) to capture the wide brightness range. Then the drone was rotated, and another three shots were taken. I did this three times. Each set of three photos was merged using Adobe Lightroom Classic to form one HDR photo, resulting in three HDR photos, each with a slightly different view. These resulting three photos were then merged into a single panorama photo, again using Lightroom, to create the final image.

Edmonds Waterfront
Edmonds Waterfront

Edit 2/13/22: If you want to see the above photo in larger size, look at my Flickr album here.