I've been developing my system for aerial surveying, no big deal, just an Iris+ with a Canon S100, and wondering if it ever makes sense to not use GCPs even if you have RTK.
The sticking point to me seems to be in the uncertainty of camera attitude. Even if you know the position of your UAV to within 1cm, if your camera angle is off by 0.25 degrees from dead vertical, it's going to be pointing at a point 43 cm from the sub-aircraft point, in some unknown direction.
Clearly the only way for a multicopter to move forward is to alter its pitch. Pitch and roll will vary depending on the wind. Pitch and roll are logged by the UAV, but attitude is not part of geotagging of photos as far as I know. Maybe it is, and this question is moot.
But if it's not moot, then it seems like you always are going to have to have GCPs. Maybe you could get away with a single GCP?
You definitely need several GCP's, I don't even geotag anymore. The RTK still needs survey equipment that goes with it. I get a surveyor to give me the XYZ's for the GCP's. I consider myself a data collector, not a surveyor or mapping expert.
In CA, a P.E. has to sign off on your final product.
Not much work has been published on this important question. I think that in the not too long future, geo-referencing by means of precise camera exposure positioning will become common practice. The placing and surveying of targets on the ground is currently the largest manual component in the image acquisition phase of a mapping project. Elimination of the need for GCPs thus has significant economic promises. Here are two reports with robust error analyses of maps made with SfM and no GCPs:
As far as the need for camera orientations is concerned, these can best, and most accurately be derived in the camera alignment step of the SfM workflow. Trying to measure them physically by means of small, lightweight IMUs is rather complicated and not nearly as accurate.
It really depends upon what data you want to create or extract as the end results, but generally speaking you will want RTK/GNSS survey grade GCPs. The more complex the terrain, the more GCP's. If you are in an area in which you can not get GPS fix's.. then some higher end photogrammetry software supports scaling/orientation, in which you can physically measure distances between two or more points (i.e.: with a tape measure), and/or use a laser level between two points, and enter this into the software for scaling/leveling... we do this for stock pile measurements when in side large buildings, or mines.
currently only pix4d supports scaling, leveling ?
As Harry Tuttle has said, in order to achieve accuracy in your model you need to have precise GCP's and by of of precise I mean measured with a GNSS sensor. More are better than few and most definitely a single GCP won't cut it. I tend to aim for 10+ GCP's per 100 acres positioned at carefully chosen locations over the area of interest to give texture and terrain height variations, covering both internal and boundary points. I don't worry so much about the dead vertical aspect as good processing software will deal with this, getting sufficient overlap is more important, having each single point on the ground visible from 6-10 images. MP will geotag using CAM messages if you are triggering the shutter using a cable, you just need to check on shutter lag though with the precise GCP's the camera positions can be optimized.
For precision surveying you will need a GCP. It is what gives exact distances from a know landmark to your unit. Most surveying crews use a RTK unit for the highest level of accuracy. If you're using an RTK and have accurate GCP's then you will have 3 point fix. A few questions I have for image processing. What are you using for geotagging? If you're using mission planner is your cameras clock synced to your mission planner. Lastly what image stitching software are you using? I recommend agisoft.
I'm just using Pix4Dmapper for now. Actually I've only done one survey, because my aircraft crashed after my IMU went out and I'm waiting for a new Pixhawk.
For my one survey, my S100's GPS didn't get enough satellites and didn't tag the images. But I had laid some GCPs out and taken photos of them with my iPhone, so I took the coordinates off of those, then manually matched them to my survey images in Pix4DMapper. I got about 8 to 11 images per GCP, and the result lined up with Google Maps fairly well.
Now I'm not sure if the survey images need geotags at all. It seems like the only thing they're good for is getting an approximate location, and then the rest has to be done with accurate GCPs and the image stitching software. If you don't have GCPs, geotags can help locate the resulting map, but it would most likely have a location uncertainty of 1 to 10 meters.
If you could know your camera's attitude accurately, then you could use geotagged images, but you'd have to know that to within 0.005 degrees to get centimeter accuracy at 100 meters altitude. So it seems like RTK is not useful for most aircraft, just for the GCPs. Unless you're using something like the eBee RTK that flies the "weird" pattern people are talking about. I guess it has to fly that way, because otherwise the RTK is for nought.