Topics

1

General / size of non coded targets

« on: February 04, 2016, 03:11:29 PM »

Dear All,
I am about to test usage of non-coded targets in an aerial mission and their automatic detection by Photoscan. In the tutorial about coded targets there is a suggestion on the size of the coded targets. But I cannot find any indications about the rage of sizes for the non-coded targets (not in the manual, neither in tutorials). Could someone shed some light on that? Any experiences?

kind regards,
Ola

2

General / point pixel coordinates in undistorted images

« on: November 03, 2015, 01:02:32 PM »

Dear All,

Based on the camera model estimated in Photoscan and coordinates of an object in a distorted image (col,row), I try to compute coordinates of the object in the undistorted image (col_und, row_und) provided by Photoscan.
I export the images without centering the principal point (but I also tried with this option enabled).
I follow the formulas provided in the user manual but cannot get the values right. There is only the radial distortion that is present in the images (all tangential distortion = 0, as well as the skew value).

I noticed that the undistorted image should be larger than the original, but the exported undistorted image is of the same size as the input image. I suspect this "crop" causes a shift in my computations but I am not able figure out what the sift is.

Could some shed some light on this subject? How to find coordinates of an object in the undistorted image exported in PS based on the camera model and coordinates of an object in a distorted image.

best regards,
Ola

3

General / 'Reference' based image alignment - images divided in 2 groups

« on: February 16, 2015, 05:24:41 PM »

Hi,

I work with a set of geo-tagged images acquired in one strip with >90% forward overlap and try to align them using the 'Reference' pair selection method. The images are nadir looking but I set the average terrain height anyway. I am aware of the fact that single image strips are far from being optimal for photogrammetric image processing, but I try to understand why the image set is always aligned in two groups (see attachment). Images within each group are aligned correctly, but in between the two groups alignment is incorrect. Initial image geo-location (GPS positions) looks logical and correct (with reasonably equal spacing between successive images in the strip), but Photoscan doesn't even look for matches between the last image in group 1 and the first image of group 2. Any suggestions on which parameter could I tune to avoid this behavior? Processing data using Generic or Disabled pair selection mode is much more time consuming....

best,
Ola

4

Python and Java API / Fisheye sensor - setting type

« on: January 29, 2015, 05:01:18 PM »

Hi All,

I need to write a simple script to automate data processing: image alignment, dense points cloud creation, meshing, etc. The one shown in page 5 of the Python API Reference works well for a frame camera, what about a fisheye sensor?
Anyone knows how to set the sensor to Fisheye?

best regards,

Ola

5

General / Hardware requirements - GPU

« on: January 16, 2014, 01:02:36 PM »

Hi,

We have a machine with 24 Gb RAM, 2 x 2.67GHz CPU and a mother board meeting Photoscan requirements but the graphic hard seems to be a weak point:

•   Graphics Processor
o   ATI ES1000
•   Video Memory
o   32 MB
•   Video Interfaces
o   VGA

(see http://www.techpowerup.com/gpudb/2102/es1000.html for more details.)

Could you please provide indications about what is the influence of the weak GPU on the performance of Photoscan in image alignment and dense point matching?

Would changing the graphic card to this one
http://www.asus.com/Graphics_Cards/GT6401GD3L/#specifications
improve the performance dramatically?

best regards,

osima

6

General / Camera based chunks alingment

« on: January 03, 2014, 03:38:49 PM »

Hi,

I have problems with understanding how the camera based chunks alignment works.

I have a project including several chunks. All the chunks include images from one camera (precalibrated, fixed), acquired from different positions. There are four images common to all the chunks (e.g.: A.jpg, B.jpg, C.jpg, D.jpg are present in every chunk).
When I select Workflow/Align chunks/Camera based alignment (fix scale unchecked), the chunks are not aligned as I would imagine they should be. In another words, after alignment the position of A.jpg, B.jpg, C.jpg, D.jpg from Chunk 1 is not the same as position of A.jpg B.jpg, C.jpg, D.jpg from Chunk 2 (they seem to be shifted).

Could someone shed some light on the camera based chunks alignment? I would like to understand what I do wrong.

Best regards,
Ola
 

7

Python and Java API / AlignChunks runtime error

« on: December 11, 2013, 02:11:27 PM »

Hi,
I am a beginner in Python, but managed pretty well so far.

At least until I got this cryptic message thrown by the console:

Traceback (most recent call last):
  File "//tsclient/D/Tools/ownScripts/Python_PhotoScan/first_test_multiple_chunks_oneCalibration.py", line 50, in <module>
    PhotoScan.alignChunks(doc.chunks,doc.chunks[1], method="points", accuracy="high", preselection=False, filter_mask=False, point_limit=40000)
RuntimeError: More keyword list entries (7) than format specifiers (5)

This is the code that I use:

Code: [Select]

#PhotoScan 0.9.1 build 1703 (64bit)
import PhotoScan
import os, os.path
import string

doc = PhotoScan.Document()

path="X:/FirstData/Chunks/"  #path to the image folder
cameraCalibPath = "CamCalibration.xml" #path to the camera calibration file
cameraCalibration = PhotoScan.Calibration()
cameraCalibration.load(path + cameraCalibPath)
print (cameraCalibration.width)

#For every folder with image subsets
for root, dirs, files in os.walk(path):


    for name in dirs:
        photos_list = os.listdir(os.path.join(root, name))
        chunk = PhotoScan.Chunk()
        chunk.label = name
        doc.chunks.add(chunk) #add a new chunk
        sensor = PhotoScan.Sensor()
        sensor.width = cameraCalibration.width
        sensor.height = cameraCalibration.height
        sensor.user_calib = cameraCalibration
        sensor.fixed = True
       
        for photo_name in photos_list:     #add all images from the folder
            camera = PhotoScan.Camera()
            camera.open(os.path.join(root, name) + "/" + photo_name)
            if (photo_name.rsplit(".",1)[1] =="jpg") :
               camera.label = camera.path.rsplit("/",1)[1]
               camera.sensor = sensor 
               chunk.sensors.add(sensor)
               chunk.photos.add(camera)

doc.save("X:/FirstData/Chunks/proj1.psz")

#Align Photos in every chunk
for chunk in doc.chunks:
    chunk.matchPhotos(accuracy = "high", preselection="disabled", filter_mask=False, point_limit=40000)
    chunk.alignPhotos()   
    #Build Geometry
    chunk.buildDepth(quality="high", filter="aggressive", gpu_mask=0, cpu_cores_inactive=1)
    chunk.buildModel(object="height field", geometry="smooth", faces=0, filter_threshold=0, hole_threshold=10)   
    doc.save("X:/FirstData/Chunks/proj1.psz")

#Align chunks
PhotoScan.alignChunks(doc.chunks,doc.chunks[1], method="points", accuracy="high", preselection=False, filter_mask=False, point_limit=40000)
PhotoScan.mergeChunks(doc.chunks, merge_models=False, merge_markers=False)



Could anyone shed some light on what I do wrong?

Best regards,

Ola