We’ve been wanting to improve the efficiency of our drone camera projects and I’ve always wanted to eliminate GCPs as on large projects can be a real pain.
On the M2 page it says the M2 allows you to have “fewer” GCPs. Can it eliminate them altogether?
We would be using a Reach RS2 for base station (most projects we don’t need global corrections, our local known point is fine to keep repeat surveys consistent). Speaking to an M2+. Which we would hotshoe into our Micasense Dual MX.
I know in the past, there were issues with the Emlid M-line and Micasense cameras but my reading suggests this is fixed.
The end result we’d want is all our photos are RTK (to our local known point, or globally if we used corrects like CORS) and we wouldn’t need to correct with base stations.
I wouldn’t recommend completely getting rid of GCP’s unless you have an established means of rectifying the drone data exported from photogrammetry (Pg). I personally do not process with GCP’s. I perform Pg reconstruction with the appropriate Geoid which gives me the mathematically correct global accuracy.
From there I rectify the exported data in either Agtek, Carlson Precision 3D or Civil 3D using surveyed checkshots. Which software depends on what the deliverable is. If all I need is the ortho and DTM points then it stops at Carlson. If I need a decimated DTM with nice smooth contours and/or need to do some earthworks calculations then it stops with Agtek Gradework. If it is the full blown integration with other design data and/or includes planimetrics then it goes all the way to Civil 3D. If you do not have any of these capabilities then you may be able to rectify the vertical in your Pg software utilizing a geoid or point adjustment but it’s less likely that you will be able to rectify horizontally to a different coordinate system’s position.
I do not use GCP’s because I work with Surveyors, Contractors, Engineers, Architects and GIS professionals and prefer to leave my native data as it comes corrected off the drone. This allows me to transform and rectify the data to any source I want without having to figure out where it was last manipulated. There’s always a source of truth.
I’ve been playing around with my RS2 and my EVO II Pro RTK now for about a year experimenting with and without GCP’s with realtime RTK either from my RS2-EVO (state RTN) or just using state RTN for the EVO.
When I first got started, Dave Pitman here on the forum really helped me to understand the intricaties of the camera calibration as well as the use of “check points” for the GCP’s in the imagery. It all finally made sense after a couple of experimental flights.
In our state we are very fortunate to have 2008 and recently 2020 lidar models. Since 2008 and even long before verifying individual photogrammetric/lidar projects, we’ve found accuracies well within the tolerances as stated in their projects. We were verifying GCP’s established by others as well as “check points” either by trigonometric traverse methods in wooded areas and/or GNSS methods of photogrammetry and lidar products.
I’ve found using lidar derived surfaces to be a good method of verification when just using the EVO RTK either by local base RTK (RS2) or just using the EVO RTK via our states RTN without GCP’s. As the lidar data is based on our state’s plane coordinate system, it’s a good check for the flown imagery data.
Of course you simply can’t use any point on the lidar surface without the selected “check point” being a hard surface that will coincide with your imagery. I’ve found you can achieve with care using the EVO RTK only method (states RTN) a 2-3 cm accuracy in both horizontal and vertical components for the derived surface. I’ve also found that obtaining adequate imagery with “overkill” is imperative in using this method. This means either using double oblique methods or double nadir methods obtaining the imagery at adequate altitudes to obtain the accuracy above.
However, in my opinion this doesn’t replace the use of GCP’s in your project. They are important in verifying your final surface derived by photogrammetric methods. I’m not a photogrammetrist… just an old professional land surveyor wanting to keep my mind active with learning new things. So far this subject has really fascinated me. Especially using Metashape software, thanks Dave ! I’ve flunked my first part 107 test… been having too much fun flying !
Michael and Dave are correct… make sure you have sufficient GCP’s for your data. If not, you are taking a big gamble in deriving your final surface .
I don’t know if it is easy to have RTK positions with RS2/M2 on a drone. How will you send the RS2 corrections to the M2 on a drone? On DJI drones the corrections are sent to the drone with telemetry from the remote control.
Otherwise you could PPK the RS2/M2 raw files to get accurate photo positions.
As the others said you must place GCP’s on the site and use them as “check points”.
Via wifi, either 2.4 or 5.8 ghz which is pretty much standard. Using the RS2 hotspot via the controller of your drone, the controller relays the RTK data stream to the uas. If you don’t have RTK available, just start the base logging as well as the M2 logging on the uas. After the flight, you can uses Emlid studio to post process the data. Emlid has excellent tutorials on this here on their website.
What kind of drone? You should be able to use a LoRa adapter for the M2 and pass corrections via radio. I’m not 100% that Mica is capable of writing via that method through the hotshoe adapter but I’m sure someone on here knows.
Brian. If you are flying the same fields over and over, why not set some control that is persistent? Then you can eliminate subsequent setting of points and still have gcp / check shots for your processing.
We haven’t flown any sites yet that require numerous visits, however I’ve put out some feelers to some construction firms in the area. I’m kind of hesitant to really advertise any uav capabilities yet until I’m faa certified, however it’ll happen soon enough. If I do get any projects, they’ll have permanent control/photo targets.
I’ve been mostly just doing some in house testing for simple planimetric mapping and topography for large boundary surveys when there’s no wooded areas. Flying some of the large boundaries, there’s not much pasture lands and the few that I’ve flown have been for the location of existing buildings and rural driveways/roads. It has really helped to reduce our workload in locating structures and roadways in the field by conventional means. It’s fascinating to compare the final surface product against the 2020 lidar, I’m continually amazed at what the Evo and Metashape can accomplish. I’ve compared hard points via GNSS/flown surface/lidar product and have hit target elevations and horizontal within 3-5 cm. It’s truly amazing what you can obtain with enough care in the methodology of flying and processing the data.
It’s also amazing the details you can achieve flying at various altitudes, especially low level (30-50m). However the photo counts go sky high at 80% forward and side overlap when flying low as well as battery use. You really have to make sure you’ve got enough batteries before the flight. Processing these with high photo counts, really takes a lot of time even on my new Dell 7865 workstation with 64gb ram, and AMD threadripper processor with a 24gb Nvidia graphic card. The last site I flew was 15 acres of farmland with a lot of structures at 30m flight level. I ended up with about 1400 images. I know it’s overkill but I’m still experimenting. The details were amazing looking at the cows and goats grazing as well as the various farm buildings.
I think the most fun besides flying is the processing of the imagery in Metashape. I found there’s all kind of tweaks you can use to really refine the final product. I’m no expert yet in Metashape processing, but I’m confident in what the final surface product is whether checking by GCP’s or lidar hard points (roads, etc). It’s also fascinating regarding the NGS Geoid18 model used when converting the processed imagery from ellipsoid to the orthometric surface in Metashape, their conversion routine is pretty accurate. I’m convinced using the ellipsoid in all imagery processing then converting to whatever datum/projection you need is the proper methodology.
Thanks again for recommending Agisoft. I’ve only begun to fully realize the software capabilities. I’ve got some ideas scanning historical buildings in the area either via terrestrial or flying and processing in Metashape. Whether or not it’ll come to any fruition for marketing will depend on getting certified.
I’m wanting a lidar system, but it’ll have to wait until I’m certified. The cost is kinda high for the limited use we have now but we’ll get a system. I see it in the horizon, I just have to get there !
When I flew the Autel my balance seem to come at 200-225ft and 75/65 overlaps. If vertical objects became too numerous and close together to allow sufficient collection of points in between I would go to a crosshatch nadir at 70/60. I found that this increased the density of the cloud enough without exploding the other factors. The reason we’re not getting enough points is because the oblique angles are blocked. If I ever run obliques they are usually isolated on the structures and then combined with the nadir cloud later but occasionally I want a nice overall 3D model and am not so concerned with terrain accuracy so I’ll run the full site oblique. I always run a nadir path. Very rarely do I run a full site crosshatch unless it is something like an urban environment of populated subdivision.
My comment was actually meant for the OP, Brian Lindens @intenna.
But it has been great seeing your evolution integrating photogrammetry into your tool kit. LIDAR is the next logical step. But for open topography, there really isn’t a meaningful precision jump over photogrammetry. Vegetated topography and infrastructure is where LIDAR can shine.
I am waiting excitedly to see more testing of the Zenmuse L2! Preferably I want a dataset in my hands. That price point would put an entire setup below $30k and that becomes a budgetable expense for our small business.
That’s a great idea Bryan that never occurred to me. You’re saying just put in permanent GCPs and the never need to remark.
The challenge will be these are farm sites, so there really is no area where the tractor isn’t driving. You’d be shocked at what these tractor drivers can manage to hit haha. But I’ll give that some thought if there’s a way.
It’d also be great to reduce/eliminate GCPs on big projects where we can’t put down many GCPs. We’ve had some mountain projects, even where we helicopter in, and so in some cases haven’t been able to GCPs at all. Small number of our projects and super accuracy isn’t critical, so haven’t worried to much about that currently.
Just because the position of the photo is more accurate, the yaw, pitch, and roll are going to be lacking. GCPs also function as tie points, which help match up locations of pixels from various photos.
I use a PPK module on the drone (from Topodrone) and RS2 as a base. I think the same considerations apply.
I place some points as check points. In my opinion they are essential. Good point is you can place check points in easy to reach areas or use natural targets where they suit you.
Photogrammetry tools I know allow you do decide if a point is control or check.
Reach M2 can output its centimeter-accurate position at regular intervals in RTK and the drone’s autopilot can use the position for RTK navigation. However, the transmitted positions don’t match with image centers. Thus, Reach M2 can’t be used for precise mapping in real time.
However, you can use PPK with your current setup. While Reach M2 can achieve centimeter-level accuracy with PPK, I still recommend using GCPs for your projects. This will ensure consistency with the results of your previous flights. It is also a good practice to place GCPs in areas with abrupt changes in elevation. Additionally, place some checkpoints during processing in Metashape to validate the resulting model.
