Messages

31

General / How does PS Pro handle geodetic datum transormations?

« on: September 15, 2013, 12:23:45 PM »

Hi everybody,

I just recently came across the question, whether PS Pro automatically takes the appropriate geodetic datum transformations into account, when conversion between coordinate reference systems requires to do so.

Example:
Photoscan project file was set up using manually picked gcp from CAD drawing which was in DHDN GK Zone 3 (EPSG:31467) projected coordinate system.
Parameters for that are (in ESRI-PRJ File compatible notation):

PROJCS["DHDN_3_Degree_Gauss_Zone_3",
GEOGCS["GCS_Deutsches_Hauptdreiecksnetz",
DATUM["D_Deutsches_Hauptdreiecksnetz",
SPHEROID["Bessel_1841",6377397.155,299.1528128]],
PRIMEM["Greenwich",0.0],UNIT["Degree",0.0174532925199433]],
PROJECTION["Gauss_Kruger"],
PARAMETER["False_Easting",3500000.0],
PARAMETER["False_Northing",0.0],
PARAMETER["Central_Meridian",9.0],
PARAMETER["Scale_Factor",1.0],
PARAMETER["Latitude_Of_Origin",0.0],
UNIT["Meter",1.0]]


Task was, to convert to geographic coordinates in WGS84 (EPSG:4326), which is defined by:

GEOGCS["GCS_WGS_1984",
DATUM["D_WGS_1984",
SPHEROID["WGS_1984",6378137,298.257223563]],
PRIMEM["Greenwich",0],UNIT["Degree",0.017453292519943295]]


Depending on the location of the project area, there are different Datum transformations applicable (here as suggested by ArcGIS projection engine):

DHDN_To_WGS_1984_1           1673        Germany - West
DHDN_To_WGS_1984_2           1777        Germany - West
DHDN_To_WGS_1984_3           15869       Germany - Brandenburg, Mecklenburg-Vorpommern, Sachsen, Sachsen-Anhalt, Thueringen
DHDN_To_WGS_1984_3x          108206      Germany - South of 50°20' N
DHDN_To_WGS_1984_4_NTv2      15949       Germany - onshore
DHDN_To_WGS_1984_4x          108207      Germany - between 50°20' N and 52°20' N
DHDN_To_WGS_1984_5x          108208      Germany - North of 52°20' N
DHDN_To_WGS_1984_6x          108209      Germany - Thuringen
DHDN_To_WGS_1984_7x          108210      Germany - Sachsen


In ArcGIS projection engine these are implemented as 7-Parameter Helmert-like transformations with the following parameters (except NTv2):

Naming               Code     Method            dx     dy    dz     rx       ry       rz     ds
-------------------------------------------------------------------------------------------------
DHDN_To_WGS_1984_1   1673     Coordinate_Frame  582    105   414    -1.04    -0.35    3.08   8.3
DHDN_To_WGS_1984_2   1777     Position_Vector   598.1  73.7  418.2   0.202    0.045  -2.455  6.7
DHDN_To_WGS_1984_3   15869    Position_Vector   612.4  77    440.2  -0.054    0.057  -2.797  2.55
DHDN_To_WGS_1984_3x  108206   Position_Vector   597.1  71.4  412.1   0.894    0.068  -1.563  7.58
DHDN_To_WGS_1984_4x  108207   Position_Vector   584.8  67    400.3   0.105    0.013  -2.378 10.29
DHDN_To_WGS_1984_5x  108208   Position_Vector   590.5  69.5  411.6  -0.796   -0.052  -3.601  8.3
DHDN_To_WGS_1984_6x  108209   Position_Vector   599.4  72.4  419.2  -0.062   -0.022  -2.723  6.46
DHDN_To_WGS_1984_7x  108210   Position_Vector   612.4  77    440.2  -0.054    0.057  -2.797  2.55


dx = x axis translation (meters)
dy = y axis translation (meters)
dz = z axis translation (meters)
rx = x axis rotation (arc-seconds)
ry = y axis rotation (arc-seconds)
rz = z axis rotation (arc-seconds)
ds = scale difference (parts per million, ppm)

(Source: ESRIS ArcGIS Documentation, ArcGIS 9.3.1, geographic_transformations.pdf)


My question is:
How are these things handled by Photoscan PRO?

Cheers

32

General / Re: Agisoft PhotoScan 1.0.0 pre-release

« on: August 27, 2013, 12:09:45 AM »

Hi,

Suggestions to make interfacing to other software more comfortable:

Import and export functions for dense point cloud / mesh to Point Cloud library PCD format and Cloud Compare Bin-Format, maintaining XYZRGBNXNYNZ attributes. Would save a lot of time that is otherwise used for intermediate conversion-steps.

Would be really great!

Cheers

33

General / Re: Agisoft PhotoScan 1.0.0 pre-release

« on: August 26, 2013, 11:46:21 PM »

Hi Dmitry,

very good and very useful new functions. I am currently playing around with it  under Linux Mint 15, 64 bit, nVidia Driver Version 310.44.

Image set of 153 21 megapixel aerial images from Canon EOS 5d

Alignment, Scene reconstruction and dense point cloud in high mode work fine (240 Mio point).
But mesh generation fails, independent of settings, in the following way:

mesh generation dialog comes up after a short while, options can all be set as necessary.

When clicking ok, mesh generation starts, when progressbar reaches approx. 10-12%, progressbar dialog is closed and a mesh with 0 faces is generated. Thats it. Am I missing something?


Cheers

34

General / Re: Problems Making DEMs from Video Frames

« on: August 17, 2013, 06:37:25 AM »

Hi,

please let me ask how you performed the georeferencing process? Enough ground control points?
There are markers for each image in the sreenshots, are these also ground control points? What CRS did you use when visualizing data in ArcMap?

Cheers

35

General / Re: Octree-Level

« on: August 17, 2013, 06:28:04 AM »

Hi Alexey,

we often observe that PS calculates Octree-level > 10 also for low and medium meshes with moderate spatial extend when working with full resolution images (ground pixel size 2 -7 cm, 5616x3744 pixel). PS tends to fill up the RAM when using arbitrary mode - which we need for a full 3D Object reconstrucion - and it tends to freeze during mesh decimation. It works in most cases with Octree-levels <=10. So we would like to have control on that "screw". Furthemore we would like to have the processing split up in single steps, calculating the dense cloud, filtering it possibly,  then meshing.

For the moment we export the dense point cloud and do the rest in third party sw, mostly PCL.


Cheers

36

General / Re: Trouble aligning, ower water

« on: August 17, 2013, 05:49:03 AM »

Hi,

the basic principle behind classical stereophotogrammetry as well as for  the 3D scene reconstruction from multiple (aerial) images as Photoscan does, is, to have homologous points - identical object points viewed from different camera positions. In classical photogrammetry this is done by an Operator, who identifies and measures the positions of these points (control-points with known world coordinates as well tiepoints between images). Photoscan manages this by automaticallly extracting potential point canditates by feature extraction algorithms (maybe SIFT, SURF or EPFL DAISY). Their charackteristics are stored in so called feature descriptores which are analysed by statistical means during matching to find matching points.

To conclude: If it is impossible for PS to find enough matching points it cannot do a proper scene reconstruvtion. And for  the following surfaces / Objects this is hard to do or nearly impossible:

- Water, because it is mostly  a moving  surfaces, when it is  not moving, it is textureless and reflects incoming light specular.
- Man made structures without any texture, specular reflective charackeristics  e.g. Metal roofs, glass surfaces, uniform concrete surfaces etc.
- Moving objects like cars on a motoway etc, ships , trains etc

As George suggested, you can use smooth and fill holes, so that the areas where there is no 3D reconstruction possible will be filled. But be aware that ther will possibly be mountains and valleys in your water area.

 We handle it the following way:

During point detection, matching and mesh generation water is masked out. when we have the mesh the masks are taken from the images and Orthofoto is calculated with  the complete images (use fillholes Option) . For your DSM you can fill up the water levels in your GIS software.

Cheers

37

General / Octree-Level

« on: August 14, 2013, 09:06:54 PM »

Hi,

when creating meshes with arbitrary/smooth  a Poisson surface reconstruction is performed and PS tries to generate a water-tight surface. One important parameter that controls the richness of detail for the surface is the octree-level. While in other software packages like meshlab, cloudcompare, PCL etc this parameter can be set manually within a defined range, PS calculates it automatically.

I would lke to know which parameters this calculation is using, e.g. regionbox size in combination with local point density or similar criteria?

I really would appreciate to have this option under manual control.

Cheers

38

General / Re: viewing 20M+ pointcloud in UTM coordinates

« on: August 14, 2013, 08:53:44 PM »

just have overlooked that in the same edit menu there is also an option to scale your data.

Cheers

39

General / Re: Terrain model out of DSM

« on: August 14, 2013, 08:47:46 PM »

Hi

there are some moving-window filter for slope-based filtering of raster-dsm. Vosselman et. al have done great work on this for lidar data. In case of flat terrain this works quite well also for photogrammetrically derived point clouds and you can get rid of most surface objects. But there is one big problem: forested areas. when there are not enough 3D points on the ground, there DTM reconstruction will be very poor or at least totally wrong and cannot compete with lidar. Other options will be to directly concentrate on the point cloud itself. May you want to take a look at PCL (Point cloud library) There are some nice filters, but I am running into performance problems when dealing with massive pointclouds of several Mio of points, which are common when doing terrain reconstruction

Please report when you have found a nice way.

Cheers

40

General / Re: viewing 20M+ pointcloud in UTM coordinates

« on: August 14, 2013, 08:25:02 PM »

Hi,
I also have not digged too deep into cludcompare yet and did not test the following:

when you select  the pointcloud in the list of data objects you can open from Edit menu a dialog to apply a transformation matrix. Try with editing TX,TY,TZ  to fit to your needs.

Hope it will work...

Cheers

41

General / Re: viewing 20M+ pointcloud in UTM coordinates

« on: August 09, 2013, 04:54:41 PM »

Hi,

you may like to try CloudCompare. It is OpenSource and free.

http://www.danielgm.net/cc/release/

CloudCompare is a nice tool in addition to meshlab and it shifts coordinates temporarely and gives you a notice about it. 2o Mio points should be possible with it.

Cheers

42

General / Coded Targets as Markers in Aerial Surveying - Exxperiences?

« on: July 04, 2013, 06:46:58 PM »

Hi everybody,

we use coded targets as standard in close range photogrammetry, for which they work also fine in PS.

But I was wondering, if they may be also usefull  - time-saving - in aerial surveying with UAS and UL aircraft up to mid altitude like 700 to 1000 metres a. Gr.

I ran some simple tests with a 12 bit scheme and came to the conclusion that the coded targets should be imaged at least with 40x40 pixel, otherwise they are not recognized by the search algorithm.

Can somebody confirm this or has other experiences?

Thanks