This originated as a blog for my trailcam images, but I have been expansive in my definition of which cameras to include. In addition to motion-triggered images, both trailcam and DSLR, I also post images remotely triggered. Including images from flying cameras, aka drones.
As with my terrestrial photography, I don't make nearly enough income from drones to support the fambly, but I get enough to report to the IRS. I have two drones, a DJI Mavic 3 Pro that I use for my contracted projects, and a DJI Mini 4K that I use for...what? Even though it weighs less than 250g and doesn't need to be registered for recreational use, I registered it with the FAA so I could use it on a project if I had to. I used it two years ago when I flew my big drone into a tree and I needed just a few more shots. (I haven't used Sport Mode since then.) The Mini 4K is one step up from a toy, but it is not a big step. To me, its primary deficiency is it does not do waypoints.
I have become somewhat proficient using waypoints on the Mavic. I was hired for a project recently that required some video in addition to stills. (I can't say what it was because I had to sign an NDA, which is a first.) Rather than shooting video while flying manually, it is much easier to get smooth video by programming waypoints and having the drone fly the program. I framed the image while flying, set a waypoint, and repeated that 5-6 times for each video. The controller shows the projected time of the flight, so I adjusted the speed of the drone to fit the video in 60-90 seconds.
One problem with doing this is the drone has to fly the route twice; once to set waypoints manually and once to shoot the video automatically. Multiply this by five videos plus some stills and I was getting nervous about whether three batteries was enough. I had the Mini 4K along as a backup, but since it can't do waypoints it only would have been useful for stills. Fortunately I made it through with about half a battery remaining.
But it got me thinking about what I really need in a backup drone, and the Mini 4K is not it. I put it on eBay, and today I am expecting delivery of a DJI Mini 5 Pro, which does do waypoints and actually may be easier to program than the Mavic. In order to plan missions on a desktop computer rather than a tiny controller screen, I've been experimenting with a couple of programs and for now have settled on Litchi Hub. It produces files with a KMZ extension, and loading to the Mavic controller is a convoluted process. Supposedly it is slightly easier to do the upload with the Mini 5 Pro controller, which is more advanced (RC2 vs. RC). DJI could make this process much easier, but they choose not to. It is apparent that they want you to buy their more expensive enterprise drones and their planning and processing software.
In anticipation of today's visit from UPS, I've been watching YouTube videos about waypoints and ended up going down the rabbit hole of orthomosaics. These are stitched images shot with the camera pointed straight down. Once you get more than a couple of acres it might be impossible to get high enough (legally) to get the entire property in frame for a top-down shot, and stiching images also avoids the distortion at the edges that a single shot would have. As I was watching the videos, I learned that there are different levels of precision with orthomosaics, and the more precise you need to be the more it is going to cost for equipment and software. (On one of the videos, Dave Young of Drone Launch Academy calls these levels: Tier 1 - Relative Accuracy, and Tier 2 - Absolute Accuracy.)
Complicating the matter further, the US government has decided that DJI and other Chinese drone companies are national security risks, and will not approve new models for use in this country. Previously-approved models are still allowed, but DJI no longer sells them directly and (more importantly) does not offer the warranty program DJI Care. Even though existing models are not totally banned, they can't be used on US government or critical infrastructure projects. There are some US-made enterprise drones that are rather expensive (Skydio, Parrot, Freefly), but as far as I know there are no US-made affordable prosumer drones like the Mini 5 Pro.
The drone I am getting today probably is a South Korea gray market item, and there is a warranty but there are extra hoops to jump through to enforce it. When I crashed my Mavic two years ago and knocked the camera gimbal out of alignment, it was covered by DJI Care. All I had to do was send it in (to Texas, not China) and they sent me a different one, probably refurbished. It cost $175 in addition to the Care cost, but I shudder to think would it have cost (in time and money) if I had to get it repaired myself. So I'm going to be careful with my new drone.
Back to orthomosaics, it is possible to do them with the Mavic and the new Mini using low-cost (or free) planning and processing software. I believe Litchi Hub is free, but if you get a drone that can be controlled directly by their software that it a one-time payment of about $26. The Mini 4 Pro was such a drone, but I don't know if its successor will ever have that capability. For processing, I've been experimenting with RealityScan, which believe it or not is free software brought to us by Epic Games, the developers of the game Fortnite. Not knowing any better since I haven't used anything else, it seems to be quite capable. At the other end of the scale, there are $20,000+ drones and $5,000+ program subscriptions, which provide the precision needed by surveyors and many other professionals. I just want to shoot a few roof projects that pop up once in a while on my job boards, not run calculations on how much volume there is in a rock pile at a mine. That's something they do with high-precision drones, I learned.
During my testing, I processed two orthomosaics from missions flown by the Mavic. The first was an assignment shooting a trailer park last year, which was the biggest nightmare job I've had. I didn't know how to use waypoints then so I eyeballed the alignment, and I did run out of battery, but I got enough images to get paid. This is about 15 stitched images. I'm very impressed with level of detail and I don't see any errors:
The second image was produced from video shot a few days ago above Spruce Lodge and my greenhouse. I did use waypoints on this, but the smooth curves that were beneficial when shooting the videos mentioned above are a detriment when trying to get ortho images. Straight flight pathes are preferable for ortho, but I didn't know how to do that yet when I set it up. With the curved paths, the angle of the images was constantly changing, which may have caused problems in processing. I told RealityScan to extract frames from the video every four seconds, and ended up with 266 source images.
I am not impressed with the stitching and the level of detail on this one. If I had expressed an opinion ahead of time, I would have said stills would be preferable to video frames, and this does not dissuade me from that view. The program choked on a couple things, especially the house and garage roofs. The first two images are from the same original with different cropping.
This image is a video frame showing what the roofs should have looked like. Supposedly there is a way within RealityScan to switch which frames it is using and fix something like this, but since this was just an experiment I decided the solution was to use a better image source next time. In other words, stills. And straight flight paths.
2:30 pm, it's here! I unboxed it and hooked the batteries and controller to chargers. It requires some setup so it is not going to fly today. I did do the KMZ upload and it is much easier than on the Mavic controller.
There are other capabilities in RealityScan and similar processing programs. This is a Digital Surface Model (DSM) with elevations color coded. The blue is the slope down into town, and the yellow/orange to the left show that the property slopes upward to the southwest. To the right (north), the yellow/orange are buildings and trees, including the apartment buildings to the north not shown in the above images. There is a way to (mostly) remove the buildings and trees to come up with a Digital Terrain Model (DTM). And there is a way to export this to Google Earth Pro and look at the 3D views. I knew nothing about any of this a week ago and have a ways to go find the capabilities (and deficiencies) of my setup and what I can do with the software.
