Modeling a bomb/mines field

[Mohabon]

Hello,

we are working in Vietnam and we need to survey a beach that is unaccesible becuase there is risk of bombs/mines from the 70's war.
Before of deciding a UXO clearance, a topografic survey / mapping is to be done.

The beach is as wide as 1 km and long 2 km and there is a road that is parallel to the beach.

In few words, i can place GCP on the road, on the safe side of the road, but i canno place any marker on the beach.

I have been noticing, using PS Pro, that if you have GCPs well spread, modeling is accurated and Z accuracy is good (5-10 cm or better), but only in the area inscribed INSIDE the GCPs polygon, said GCPs polygon = polygon obtained by connecting all GCPs vertexes.

The model outside the GCPs polygon quickly becomes unreliable and at 100m or less, Z error is already 3-4 m if not more.

Now question is: is there any chance i can obtain a decent DEM in the beach area by placing GCPs outside that beach and only on 1 side? My target is to obtain Z accuracy = 20cm more or less.

Check attachment please ancd see my workaround:
1) i could place many GCPs in the safe area (left of the road) for about 1 km wide area
2) i can perform an acquisition with UAV for the entire area with overlap 90% between photos
3) I can accurately survey many GCPs, but only in the safe area
4) If i use PS Pro, at what distance from the road in the "Bomb area" i could obtain a decent Z accuracy = 10-20 cm?

Following my experience, it looks like that just at 100m on the right of the road there is big error.

Is it possible to improve somehow and have a decent mode upto the shoreline?

Thanks and regards.

[Wishgranter]

Im not expert in thi sfield BUT can use http://en.wikipedia.org/wiki/Total_station and measure some easy idetifieable points of interest there, or just measure, it should be possible to combine the points in safe zones and the one measured inside, without direct touch......

[Alexey Pasumansky]

Hello Mohabon,

As Wishgranter has already mentioned you can use Total station to measure some additional points in the unsafe area.

Another possibility is to use high precision on board GPS and combine flight log with GCP data for model referencing.

[RalfH]

I imagine that, on a beach, it could be very difficult to find a target to measure with a total station from a distance of several hundred metres. What about using the high water line as a proxy for elevation? I don't mean the actual water line which changes rapidly with the waves, but the line formed by maximum water extend at the last tidal maximum. This line should be well visible on high-resolution aerial images, and it could be possible to pick points along that line which could be used as "artificial" GCPs.

Suggested workflow:
- build model from images and known (measured) GCPs
- in model, pick points along high water line, extract their modeled XYZ coordinates
- leave their XY coordinates as they are (assuming that you care less about XY than about Z)
- apply an average (or known from high water mark measurement) height value to these points
- use the resulting XYZ coordinates as "artificial" GCPs together with the known GCPs to build a second model

I know this is only a workaround, not a perfect solution, but it might help you to improve your model. Of course, the high water line is susceptible to beach slope and offshore bathymetry (influencing wave runup), but if these are homogenous along the beach and (perhaps more importantly) the high water line was formed under low energy conditions (small waves), it should be a good approximation.

[fpbv]

Mohabon

I would use both data, gps from flying AND the total station with mirrorless target.
I think will be the best way to use.

[Mohabon]

I imagine that, on a beach, it could be very difficult to find a target to measure with a total station from a distance of several hundred metres. What about using the high water line as a proxy for elevation? I don't mean the actual water line which changes rapidly with the waves, but the line formed by maximum water extend at the last tidal maximum. This line should be well visible on high-resolution aerial images, and it could be possible to pick points along that line which could be used as "artificial" GCPs.

Suggested workflow:
- build model from images and known (measured) GCPs
- in model, pick points along high water line, extract their modeled XYZ coordinates
- leave their XY coordinates as they are (assuming that you care less about XY than about Z)
- apply an average (or known from high water mark measurement) height value to these points
- use the resulting XYZ coordinates as "artificial" GCPs together with the known GCPs to build a second model

I know this is only a workaround, not a perfect solution, but it might help you to improve your model. Of course, the high water line is susceptible to beach slope and offshore bathymetry (influencing wave runup), but if these are homogenous along the beach and (perhaps more importantly) the high water line was formed under low energy conditions (small waves), it should be a good approximation.

This is really an intelligent suggestion as long as the tidal curve is well known with reference to the Mean Sea Level. That is unknown, but it is not expensive to install a tide gauge in a safe area and use these datas to find the tide observed at the time of the flight.

On the other side, i would like to know why the PS Pro modeling algorytmh looses stability so quickly when going out GCPs polygon. If for example i would survey a square area 1km by 1 km with many GCPs INSIDE, shoulnt i assume that Z accuracy is stable also extrapolating the model by 20% = 200m? And what if i make a model as wide as 5km by 5 km?

Thanks to people answering me...

[RalfH]

I don't see the need for a tidal gauge - just measure the elevation of the high water line in a safe area of the beach. You don't need to know the elevation of the water line at the time of image acquisition, just the elevation of the last maximum transgression which is visible in the images (see attached image; didn't find high-resolution along the coast of Vietnam, so I just arbitrarily grabbed a picture from Australia). Measuring at a few different spots will also give you a rough error estimate.

The problem with the Z accuracy would bother me as well, and 3-4 m at a distance of 100 m really is a lot. Could it be a camera issue? Assuming you are using a UAV, is it a small point-and-shoot with poor image geometry? Or are you using an extreme wide-angle lens which is not modelled well by PhotoScan? Or could it be that the beach surface is too homogenous for the algorithm to correctly match images?

[macsurveyr]

It may be possible to do what you want with photos, but it is a function of the quality and consistency of your images and camera, as well as the quality and completeness of the geometry of the images. At present, using only PhotoScan would be tricky and difficult to assess and quantify the results.

Tom

[mwillis]

I suggest using the internal IMU and GPS values for the triangulation.  If you fly the entire area at 3 or 4 different elevations you'll increase you accuracies by a lot.  Compare those results again Bing or Google Maps and see how close it looks.  If it is off by much, you can try geo-referencing the results against the aerial data you do have.

-Mark

[RalfH]

Mark, as I understand it the problem is not so much the error of XY coordinates (which could be corrected using georeferenced aerial or satellite images) but the error in Z direction. There seems to be some warping going on which I suspect to be due to poorly modelled image geometry. Also, depending on where you are and what data they used to rectify the satellite images, Google Earth can be off by 10 metres or more. Standard GPS can be several metres off as well, especially in the Z direction. Of course, using many measurements one can hope that the error will be cancelled out, but I don't think Photoscan has a tool for this. Building a model from images only and then using a tool like Java Graticule 3D on the point cloud (with the GPS data for the camera positions) could perhaps help in this, but only if there is no significant warping in the model (which I suspect). So unless Mohabon is using differential GPS, they will need GCPs.

[Mohabon]

Differential GPS is not possible on cheap UAV 

[RalfH]

That's what I thought. What camera are you using? Are you trying to identify bomb craters or only generally mapping topography?

[mwillis]

Another option, if you have access to Leica Photogrammetry Suite and  can fly some high altitude missions over the entire area, you could then use LPS to generate coordinates of tie points in the UXO zone.  The tie-points could then be used as GCPs in Photoscan.

-Mark

[Mohabon]

Actually we can use cheap UAV like Sensefly SMARTCAM with 16MP camera or TRIMBLE X-100 with !"MP Camera.

The bomb/mines field is really flat and sandy, so there are not craters or anything.. It was a deterent mine fields, not a war area in 70s.

[omer]

Another option, if you have access to Leica Photogrammetry Suite and  can fly some high altitude missions over the entire area, you could then use LPS to generate coordinates of tie points in the UXO zone.  The tie-points could then be used as GCPs in Photoscan.

-Mark

Hello Mark,

How can we use LPS with PhotoScan? Do you have a workflow? I tried several times and couldn't succeed it.