Accuracy comparison project

[S&S_Shovelhead]

Hello all, to finalize my bachelors degree in Geomatics I am required to complete a large project of my choosing. My outline is as follows, any input, tips or suggestions would be much appreciated.

A COMPARISON OF ACCURACY AND OPERATIONAL CONSTRAINTS OF UAV SURVEYING VS. TRADTIONAL SURVEYING METHODS

The purpose of this project is to compare UAV (unmanned aerial vehicle) surveying with traditional methods, such as data collecting with RTK GPS and/or a total station. I will compare the relative and absolute accuracy that are achievable and the advantages and disadvantages of UAV surveying. I will look at what kind of deliverables can be produced with UAV surveying; what the cost of operation is including man hours to carry out the field work as well as the processing of the data back at the office; what the initial cost would be to bring UAV surveying into an already operational surveying firm; what applications are UAV surveying methods appropriate and beneficial for compared to traditional methods; what hardware and software are required, and the environmental and legal hurdles that challenge the operation of a UAV.


I will be using my DJI Phantom 4 Pro which is equipped with a 20 megapixel camera for capturing the aerial images. I will also be using my Nikon D5300 24 megapixel DSLR equipped with a prime lens for capturing terrestrial images if required based on the project site. The handheld DSLR will allow me to map underneath overhangs and tree cover. I will be using Agisoft Photoscan Professional to process the images from both the UAV and DSLR simultaneously. I have a PC that was specifically built to handle the demanding requirements of Photoscan that I will process the data with. I will use Autocad Civil 3D for any drafting and cut/fill calculations that are required. Sketchfab software will be used to upload my finished 3d model to the internet so the client can instantaneously view it and take scaled measurements directly on the model anywhere in the world.


I have access to Trimble R10 GPS units and a Leica 405 that I will use to carry out the traditional surveying methods.

 

[S&S_Shovelhead]

I wanted to go into depth a bit more of how I plan to carry out the project. My overall goal is to determine under what circumstances it would be beneficial to carry out a survey using a UAV vs traditional methods. I will perform the following comparsions;

1. Cut/fill - I will compare the volume of a specific area determined by the UAV using surfaces (natural ground vs. specific plan grade) that I will import into Autocad vs. a industry standard grid (10m x 10m) picked up using RTK GPS. The control will be a much more dense grid picked up with a total station and a flat bottomed pole. This will reduce the error cause by the GPS polepuncturing the surface of the ground, the poor vertical accuracy of RTK and the missed detail of the surface that is cause by the more sparse grid.
   
2. Linear dimensions - I will compare the linear distance of definable features (side of a building) that I determine from the UAV made mesh measured in Sketchfab vs. the inverse of two corner points shot in using horizontal angle offsets with a total station and o/s shots with RTK GPS. The control will be the average of several shots using a Leica Distomat
3. Relative Accuracy - The residuals of the GCP's will give me an representation of my relative accuracy
4. Absolute Accuracy - I will leave a few GCP's out of the calibration process so I can compare their coordinates with those derived with RTK. The control will be observations made on both faces using the total station.
5. Time - Compare the total time, field and office (photoscan computing time won't count) of both methods.

 

[Florida Drone Supply]

This sounds like a very interesting project and you are going into it with the proper equipment and resources. If we can help with anything please let us know.

 

[Pipeliner_USA]

I wanted to go into depth a bit more of how I plan to carry out the project. My overall goal is to determine under what circumstances it would be beneficial to carry out a survey using a UAV vs traditional methods. I will perform the following comparsions;

1. Cut/fill - I will compare the volume of a specific area determined by the UAV using surfaces (natural ground vs. specific plan grade) that I will import into Autocad vs. a industry standard grid (10m x 10m) picked up using RTK GPS. The control will be a much more dense grid picked up with a total station and a flat bottomed pole. This will reduce the error cause by the GPS polepuncturing the surface of the ground, the poor vertical accuracy of RTK and the missed detail of the surface that is cause by the more sparse grid.
   
2. Linear dimensions - I will compare the linear distance of definable features (side of a building) that I determine from the UAV made mesh measured in Sketchfab vs. the inverse of two corner points shot in using horizontal angle offsets with a total station and o/s shots with RTK GPS. The control will be the average of several shots using a Leica Distomat
3. Relative Accuracy - The residuals of the GCP's will give me an representation of my relative accuracy
4. Absolute Accuracy - I will leave a few GCP's out of the calibration process so I can compare their coordinates with those derived with RTK. The control will be observations made on both faces using the total station.
5. Time - Compare the total time, field and office (photoscan computing time won't count) of both methods.

Any updates on your work? I am a civil engineer and also do lots of surveying. I also have a drone company. So, this interests me.

 

[S&S_Shovelhead]

Found a site and field work begins on Sunday. I will continue to uodate this thread.

 

[jack1144]

Yes excellent.
You may be aware of this already, Haag Engineering has done some testing of measurement accuracy compared to actual measurement done traditionally. Not the same as what your doing but they had very good results.
link to an article
Drone photogrammetry test: Are automated 3D roof measurements accurate enough? - SPAR 3D
Thank you for doing this project

 

[vgo195]

A lot of people have done these comparisons and the results are as good as RTK GPS (1-2cm horizontal accuracy, >3cm vertical accuracy). Of course it depends on a lot of factors (GCP's , picture quality, flight height etc).
I am also doing a comparison with RTK-GPS since I am new to drone mapping and have to do the checks I need to be able to trust the technology.
I am looking forward to hearing your results too.

 

[S&S_Shovelhead]

A lot of people have done these comparisons and the results are as good as RTK GPS (1-2cm horizontal accuracy, >3cm vertical accuracy). Of course it depends on a lot of factors (GCP's , picture quality, flight height etc).
I am also doing a comparison with RTK-GPS since I am new to drone mapping and have to do the checks I need to be able to trust the technology.
I am looking forward to hearing your results too.

Do you have any links to these comparisons you speak of?

 

[vgo195]

I have some papers but they are in Greek language (I am from Greece). Also some colleagues (Surveyors) have done comparisons and come to the conclusion I told you.
Here are some papers in English to check.

 

[griter34]

I was able to achieve .01 accuracy for horizontal and vertical accuracy by using ground control points.

 

[doktorinjh]

I was able to achieve .01 accuracy for horizontal and vertical accuracy by using ground control points.

If you're seeing this with independent check shots, then you should patent your method and sell it to every manufacturer out there! I'm guessing that this is the model correction to the GCPs and not an actual check on the model itself. We have independent checks shots on every project that we fly and the general consensus is that you're able to get 1-3x your GSD in horizontal accuracy and 3-5x your GSD in vertical accuracy. The reason for this is that you're trying to pick the middle of an aerial target by zooming in on the X or center of the target. As soon as you move away from that target, your resolution starts to degrade and now the discrete center of that target is comprised of several pixels and you select one of those pixels for the program to call the "center". The higher your GSD, the more slop that you can expect in picking the exact center of the target.

 

[doktorinjh]

Do you have any links to these comparisons you speak of?

I've got a few papers that I could pass along. PM with an email and I can send you the PDFs.