Forum

Show Posts

This section allows you to view all posts made by this member. Note that you can only see posts made in areas you currently have access to.


Messages - Pixel UAV

Pages: [1]
1
General / Re: RAM vs SWAP - can the job be done of you have enough swap
« on: January 12, 2016, 03:06:05 AM »
we have only 64gb of RAM on our processing computer, but use a 256GB SSD dedicated to SWAP.  Works really well and is cheap.  We process jobs of 2000 24mp photos with no problems.

2
General / Re: Simple "stitching" after alignment
« on: July 31, 2015, 07:05:56 AM »
Hi,

Yes you can produce a overview after aligning the photos. Do do this you have to make the model/mesh first.  It does appear odd that the building of the model/mesh takes you so long.  I would expect from a 15min alignment, that the model/mesh would produce in under a minute

3
General / Re: 3D Scanning a Mountain
« on: June 16, 2015, 08:24:32 AM »
I have a few uses for scanning large mountain areas

-We have one volcano in this region of the world that has a lake in the crater at the top. Every now and again the volcano pushes up and spills the lake over the crater edge.  Accurate models of the mountain have been made to model how this resulting Lahar(volcanic mudslide) will travel down the mountaIn and what areas will be effected.

-A ex lecturer of mine was using photogrammetry to model snow levels on mountains so power companies producing Hydro Power could predict how much water they would get in the spring melt
The same professor  also measured mountain heights of our Alps using photogrammetry and them proved his results with conventional GPS
http://www.otago.ac.nz/surveying/staff/otago027774.html

-there are some government bodies that are interested in scanning large areas of land for modeling coastal erosion.  Done correctly photogrammetry can produce results quicker than LIDAr(or so i am lead to believe)

4
General / Re: Trouble with getting good Orthophoto
« on: April 15, 2015, 06:09:54 AM »
i ran this through photoscan and got slightly better results when i made the mesh in arbitrary mode instead of height field.  It stopped the tin forming from the edge of the tower to the ground which makes the jagged edges in the smaller tower. 
still need more pictures that capture the side of the tower though
https://www.dropbox.com/s/7rw9w4l1x8bq0eu/test%202.jpg?dl=0

5
General / Re: 16GB or 24 GB RAM for Photoscan
« on: April 14, 2015, 08:49:02 AM »
A IT support man who built the computer set up the swap, so i don't know how it is done
However our setup is:

Processor-  i7-4930k @ 3.7GHz

RAM - 64gb (not sure on speed but its over the standard 1600)

Graphics- GeForce GTX 770 (two of these)

SSD1 -220gb (dedicated only to SWAP space, however this is only 64gb full, sorry i'm not sure how this works, but you may get away with a smaller drive)

SSD2 -220gb(the "c" drive, photoscan and other programs installed here, if you want to save money i would ditch this drive in favor of a traditional hard drive)

Hard drive-1tb (used for storage of our jobs)

Motherboard - X79-UP4 (we got this one as it is capable of handling 3 graphics cards in case we want to upgrade with an extra in the future)

The whole system is liquid cooled as 10 hour runs on big jobs can produce some serious heat(our old machine would hit 75 Celsius which i'm told is not good.)

We will process 2000 photos of quarry surveys etc in under 15 hours without the need to split into chunks. Over 3000 photos i will split into chunks.  Our camera is a Nex7 24mp, photos are typically taken with a GSD of 1.5cm.
 I do not promise that you will get they same speeds with the same machine.  Iv been at this for two years and found that a lot of getting speed out of photoscan is about the settings and what you want for an end result. it is very easy to overdo the processing with way to high settings.

What kind of price are you hoping for?

6
General / Re: Agisoft Vs Competitors
« on: April 13, 2015, 07:47:32 AM »
Ill put my two cents in.
I am a surveyor
I have never used Pix4d (but surveying friends do)
My favourite features that photoscan does well are
-easy to find and place GCP's (Pix4D sucks at this)
-creates very good meshes for a final result.  A very good way to import into survey packages as you don't have so many points to deal with.  I believe Pix4D does this but can only make a grid of points so you loose detail(although they are working on improving this when i last checked).  Photoscan creates a proper mesh with extra point density where needed
-a very customised work workflow.  Once you understand the setting and parameters you can tune your workflow for best or fastest results very easily.  The parameters can also be changed half way through, unlike Pix4D

How that helps

7
General / Re: 16GB or 24 GB RAM for Photoscan
« on: April 13, 2015, 07:33:16 AM »
We do aerial surveys of 2000 photos or more at a time.  Have a 64gb machine with a 220gb Solid State Drive and a SWAP overflow.  A SWAP overflow is a part of your hard drive dedicated for the RAM to fill once it reaches capacity.
A relatively cheap way to boost your RAM, however will never be as fast as traditional RAM. Works a treat though

8
General / Re: Photography Tip!
« on: April 09, 2015, 08:01:19 AM »
i may be able to help with the camera resolution and the quality of the model
It is all about the size of the pixel on the cup, known as Ground Sampling Distance,GSD, by aerial photographers. 
Since you have a high resolution camera and are taking your photos from a very close range, im guessing no more than 30cm away, the size of the pixel in real life is going to be very very small, under 0.1mm i would guess.  When photoscan makes the dense point cloud the quality you choose dictates the density of the point cloud based on how many pixels will be used.
Ultra High:Every pixel
High:Every 2nd pixel
Medium: Every 4th pixel
low:Every 8th pixel
Lowest: Every 16 pixel
i.e. If my ground sampling distance was 1mm and i choose medium quality, then the dense could would use every 4th pixel in a photo to produce its dense cloud.  therefore i would have a point every 4mm
the lower quality images would produce a better result because the GSD is lower so smooths the models out. if the GSD is to high then you end up with a lot of "noise" in your models as there is always a certain amount of error in every point, which becomes very apparent when you have a lot of points very close together.

My suggestion to you is to decide on how far apart you want the points on your cup model, then adjust your photo resolution to be 4x better than what you want.  This will give you room to play when you produce your dense point cloud.  There are GSD calculators our there.
 I hope this makes sense and note that massive resolution cameras are not the be all and end all of photogrametry.

9
Bug Reports / Ghost points in Mesh from Dense Point Cloud
« on: April 09, 2015, 06:58:47 AM »
Hi All,
I recently produced a Mesh from a Dense Point Cloud.  Before producing this mesh i had deleted parts of the Dense Cloud that wouldn't be required in the mesh.  However when i created the mesh photoscan still created points for the mesh where the  Dense Cloud had been deleted. 
I thought that since i was creating the Mesh from the Dense Point Cloud that i would only get a mesh where there was points.
I have included photos to illustrate the problem

10
Hi Dominic

Splitting your output into blocks helps improve processing times.  How many photos are in your dataset, what pixel size are you setting and how long is it then taking?

Cam

11
General / Re: Control Density and Resulting Errors
« on: September 18, 2014, 08:29:16 AM »
Will this do?

12
General / Control Density and Resulting Errors
« on: September 18, 2014, 08:01:49 AM »
Hi Everyone

I have been doing some testing of control placement to see see how it affects surface models. 
Attached is an isopac of Surface A and Surface B.  The red dots are my control marks, placed by RTK GPS.  Surface A used all the control marks and Surface B only used the ones in the corners.  As you can see in some places the differences are around 200mm, I was expecting the two surfaces be a lot closer than this. 

I am aware that some of this area is outside my control, but i am only interested in the area within my control area, I have left this in purely for interests sake. 

Processing stats ETC.
Area Description - Flat paddock with trees around the outside
Photos - 524
Resolution - 24 MP
Alignment - Medium, 15,000 point limit
Control - Added all markers, optimized, tweaked points and optimized again
Dense Point Cloud - Medium, Aggressive depth filtering
Mesh - LOW from dense point cloud

Does anyone know what is causing this or have come across this before?

Cameron

13
Feature Requests / Re: lasso selection
« on: August 28, 2014, 12:03:45 AM »
+1

bit late on the topic, but i would really like to see this feature.  would make cleaning up faces a whole lot easier.  Currently i am just rotating the model and using the square select tool.  slow and annoying

14
Feature Requests / Re: Point Cloud classification by color
« on: July 08, 2014, 03:38:56 AM »
+1

This would be very useful, as we spend a lot of our time removing vegetation from our surveys.  Could it be combined with the current classification system?  So that trees may be removed based on their color and relative height to its surroundings.

15
Bug Reports / Re: Model export as dxf
« on: June 10, 2014, 02:39:07 AM »
Hi Alexey,

I recently have come across this problem as well.  I have found that AutoCAD can handle the .dxf but many other programs can't, such as BricsCAD and 12d.  Is it possible to that Photoscan could export in the old format of .dxf so that all programs can handle it?

Cheers,
Cam

Pages: [1]