H520 Mapping Accuracy

[Tonyhiffman]

Hello All,
I have an accuracy problem with the mapping output of my H520. The issue has been discussed in older threads, but they were more than a year ago, and I'm not sure if there has been additional information brought forth about it.
This issue is the accuracy I'm getting when the H520 images are processed in Pix4D. I'll layout the exact scenario, and I would be happy to hear any insights or help.
I use the drone for mapping and surveying various sites. I'm a professional land surveyor, and I generally use the drone for mapping a larger area than what I survey. I layout the ground control, a minimum of 4 points, at the outer limits of the site surveyed (ideally).Sites are usually 1-3 acres, sometimes more.
I just used it on a site recently, and I will lay out the parameters I'm using in datapilot. I set up the project to fly 200 feet, with 65% front and 75% side overlap. I hover and stop at each picture, usually flying at 15 mph.
When I bring in the images into Pix4D, I utilize the "Rolling Shutter" option of the camera. When I set up the processing, I check the "All Prior" option in the initial processing, and then generate the ortho image and 1 foot DTM contours.
In the processing report, the GCP's are generally less than 0.2 feet horizontally, but the vertical accuracy is never better than 0.3 feet on the points. The contours are all over the place when I bring them into Civil3D, varying from 0.5 feet to more than 3 feet off the survey points I gathered.
On the same site, I tested it with my Phantom 4 Pro V2. I use the same flight parameters (height and overlap), same control points. I bring it into Pix4D,and the GCP's are spot on, and the contours match everyting very well.
Is this the best I can expect from the Yuneec? I bought it because I got tired of the geofencing nonsense of the DJI product. I will reach out to both Yuneec and Pix4D, but figured I'd ask the experinced pilots here before.

 

[paternoster]

First thing I can tell what doesn't much make difference in the result is that you make the H520 to stop with every picture. It is just a waste of batterylife, better let it fly but just in a moderate speed. I have been testing mapping capabilities with H520 & E90 and I am getting into a conclusion that despite having great resolution the E90 lacks sensitivity which shows exactly as you have experienced. I have been doing tests with both the Yuneec Datapilot and with Pix4DCapture and with both of them I can produce as good orthomosaics as you can expect with drones. However if vertical information is needed I feel only way to get good results is to take oblique shots also for the same project and preferrably take these shots also in 90 degree offset or double grid as they refer to it in Pix4DCapture. The lack of sensitivity shows in as quite big noise in pointclouds, even in flat surfaces with color and materials which should produce quite flat and solid point sets. By using both nadir and oblique pictures and good RTK-surveyed gcp's I have got the vertical precision somewhere around 0.2 to 0.3 feet; although these test have been done with small areas and flown quite low, not more then 40 metres above ground. Pointcloud produced only with nadir shots is close to truth with points close to same levels with gcp's but points much below or higher like rooftops only in 2 story houses can well be something like 3 to 5 feet away from the truth.

 

[Tonyhiffman]

Thanks very much! What oblique angle do you use when you do the additional flights? I haven't tested the Pix4D capture yet, (I installed it but haven't tested it on the android yet) I will try an oblique pattern also to see if it improves. It is unfortunate though that the Phantom produces such better results.

 

[paternoster]

I've used 45 and 60 degrees, something like 60 deg might be safer if you shoot higher and/or you shoot in a place with much vertical differences. This because you don't want the horizon to get into your shots. What comes to accuracy of results those differences in angles haven't made difference I could have noticed.

 

[Tonyhiffman]

Thanks. Last question , have you tried changing any of the camera setting (exposure or anything)? Thanks again

 

[RPR]

Thanks very much! What oblique angle do you use when you do the additional flights? I haven't tested the Pix4D capture yet, (I installed it but haven't tested it on the android yet) I will try an oblique pattern also to see if it improves. It is unfortunate though that the Phantom produces such better results.

The rule of thumb, when flying the perimeter of your structure, take images every 5 meters, at between 60 and 50 degrees down, looking towards the imaginary center of the structures, so fill any holes during meshing process.

 

[RPR]

You can also do this via mission hub (through the app) but depending if you have enough space in between your object to complete the flight plan. I’d stick with manually taking the photos and stitch them in point cloud.

 

[Tonyhiffman]

Thank you. I'm mostly interested in 3D terrain mapping, I haven't gotten in to building or structure scenarions yet

 

[RPR]

Doesn’t H50 datapilot is the same platform layout as UgCS where you can see the an illustration of your flight route? If so terrain would be an easy 3D Mission.

 

[Tonyhiffman]

The datapilot is a great mission planning software, works great. The issue is the vertical accuracy of the mapping products produced from the E90 camera is pretty poor. The imagery itself is fine, but the digital output (contours, etc) is pretty poor. I'm looking for insight on how to improve it

 

[RPR]

I’m not a PIX4D user, but is there a camera preset specific for the E90?

 

[Tonyhiffman]

There is a preset, yes. One of the reasons I went with the Yuneec was because it is supported by Pix4D mapping and capture.

 

[arruntus]

Thanks. Last question , have you tried changing any of the camera setting (exposure or anything)? Thanks again

Very interesting question, I also use Pix4D. Beware of the parameters you normally have set for photographs. Some people remember them, like the aspect ratio (I usually make them at 16:9, but in photogrammetry it's better to use the whole sensor so it's better at 4:3). I leave the color scheme in the raw mode, but it's still interesting to know how others do it and if they have noticed improvement with the enchance mode or some other. It is advisable to review the parameters before undertaking a photogrammetric flight. In which version will this be corrected?

The white balance, I also usually leave it blocked. For this first I take off, I make the change and then I launch the mission.

It is also interesting to have deactivated the distortion correction option, it is advisable to have it deactivated for photogrammetry. (I always have it deactivated, but some lenses, sin of failure and in those cases for the rest activated).

Today I was just talking to a friend about that, with what inclination to do the cross missions? The last flight was 70 degrees, but he thinks 60 would have been better. I think that the inclination depends on the height, what it's about is capturing the lateral details of objects, buildings or whatever. I personally at 50 m height 45 degrees, but if it's higher it's still better to lower the tilt a little, to 60 or 70 degrees.

Remember that in photogrammetry the vertical error is up to 3 times the horizontal, so I do not see so farfetched the data you give. Another thing is that it does not adapt to reality and as Paternoster says, GCPs to correct it.

The profile of the E90 that has Pix4D, I spoke with the people of Pix4D, in principle it was for the initial model. Yuneec then made modifications to the camera, several times, and it is still the E90. That's why we have to apply the changes manually, because not even Pix4D is able to make a correct profile for this camera, the quality is not uniform among all the E90. Even so, you already know that Pix4D, the more times it processes the data from the same camera, the more it refines the parameters, it does it automatically, self-learning.

I can't think of anything else now, let's see if we all come up with an optimal configuration

 

[Tonyhiffman]

Thank you. Lots of information here arruntus. While I agree with you about the control differences (vertical accuracy versus horizontal) the problem I’m facing is that the generated contours don’t match the rest of the surveyed points at all. Its actually more efficient to just survey the entire site and use the Yuneec to create a nice orthophoto. Sounds like utilizing both oblique and nadir patterns may help

 

[Tonyhiffman]

If anyone cares
I have figured out what’s going on...
The Yuneec mapping app creates plans with true nadir (90 degrees) flight plans
Using the Pix4Dcapture software it DEFAULTS to 80 degree oblique flight plans
So I created an oblique flight plan on a 5 acre site (70 degree) double grid lots of overlap plan with the Yuneec. It matched the phantom contours PERFECTLY

I believe the cameras on the lower end drones (phantom and H520) have to have the oblique settings to get good accuracy for true contour mapping
It’s worked on a few sites now

 

[RPR]

If anyone cares
I have figured out what’s going on...
The Yuneec mapping app creates plans with true nadir (90 degrees) flight plans
Using the Pix4Dcapture software it DEFAULTS to 80 degree oblique flight plans
So I created an oblique flight plan on a 5 acre site (70 degree) double grid lots of overlap plan with the Yuneec. It matched the phantom contours PERFECTLY

I believe the cameras on the lower end drones (phantom and H520) have to have the oblique settings to get good accuracy for true contour mapping
It’s worked on a few sites now

What a coincidence, because I have been working on a simple structural mapping, and has found the same result as you reported.

6.30 Acre
210 Altitude
Flight time 19mins
2 batteries
98 waypoints
Nadir at -90°
Oblique at -77°
Front ratio at 65%
Side ratio at 65°
Ground station: GS Pro and I’ve matched similar findings flying DD.
Mapping platform: Recap Pro.

Then, I reverse or flip the grid during the 3rd and 4th battery.

I started with 80°, but what I have found interesting is when you’re flying low, with a slower speed and when my gimbal is at tilted at -77 is where I have been getting the most detailed structural (flyby) map. Not a true oblique, but it is more of a terrane mapping, because a true oblique is flown in 3 different altitude heights, followed by three gimbal tilt 80°,45°, 0°

 

[Panomapic]

If anyone cares
I have figured out what’s going on...
The Yuneec mapping app creates plans with true nadir (90 degrees) flight plans
Using the Pix4Dcapture software it DEFAULTS to 80 degree oblique flight plans
So I created an oblique flight plan on a 5 acre site (70 degree) double grid lots of overlap plan with the Yuneec. It matched the phantom contours PERFECTLY

I believe the cameras on the lower end drones (phantom and H520) have to have the oblique settings to get good accuracy for true contour mapping
It’s worked on a few sites now

Do you mean fly the pattern perpendicular to the first with a 70 degree camera angle? If so, do you fly the first one at 90 degree camera angle? I've started running into this myself so this could potentially help me a lot!

 

[Tonyhiffman]

Here’s what I’ve discovered through real life project testing. If you want reliable survey grade contours
Use the Yuneec mapping program on a double cross grid setting. At 200 to 250 feet AGL Set the camera angle at 70 degrees Front overlap 70 degrees side 70 to 80
4 or 5 good ground control points I use pix4d to process and have gotten excellent results flying with both pix4d capture and the Yuneec program
There’s no need for double nadir and oblique flights because the issue is not getting enough coverage.
I process with the extra rolling shutter option in pix4d