Topics

1

Python and Java API / Loading camera extrinsics and intrinsics

« on: July 10, 2018, 03:38:55 PM »

Hi,

I'm trying to import camera's of which I know the location and orientation.

My camera calibration file "extrinsics.txt" contains X,Y,Z and Pitch,Yaw,Roll and looks like this:

160865296_0_0.png   202228   503600   1620.77   -4.71239e-05   -0.000553269   1.57166
160965118_0_0.png   202233   502477   1618.94   0.000794125   0.000572468   4.70667
160865298_0_0.png   202651   503600   1621.07   0.000101229   -0.00047473   1.57178
160965116_0_0.png   202657   502474   1618.79   0.00010821   0.000572468   4.70528
160865300_0_0.png   203067   503601   1620.81   0.000277507   -0.000298451   1.57371
160965114_0_0.png   203074   502472   1618.99   -0.000699877   0.000486947   4.70772

Per camera I have a calibration file like this:

<?xml version="1.0" encoding="UTF-8"?>
<calibration>
   <projection>frame</projection>
   <width>17310</width>
   <height>11310</height>
   <f>16750</f>
   <cx>8655</cx>
   <cy>5625</cy>
</calibration>

I Try to combine this information with the following script:

Code: [Select]

import os
import PhotoScan
from glob import glob

# settings for this project
ext_path = "camera_extrinsics/extrinsics.txt"
img_dir  = "images_no_subs/"
int_dir  = "camera_intrinsics/"
out_path = "photoscan_out/project.psx"

# new project
doc = PhotoScan.app.document

# new chunk
chunk = doc.addChunk()

# load extrinsics
# label used in file is equal to img_path
chunk.loadReference( ext_path )

# load images and intrinsics
img_list = os.listdir( img_dir )
int_list = os.listdir( int_dir )
for img_path, int_path in zip( img_list, int_list ):

# load image
# camera = chunk.AddCamera()
camera = chunk.addCamera()
camera.open( os.path.join( img_dir, img_path ) )

# link extrinsics to image by making the label equal
camera.label = img_path

# load intrinsics
calibration = PhotoScan.Calibration()
calibration.load( os.path.join( int_dir, int_path ) )
sensor = chunk.addSensor()
sensor.user_calib = calibration

# link intrinsics to image by setting label equal (perhaps not necessary?)
sensor.label = img_path

# link intrinsics to image by assigning to camera
camera.sensor = sensor

# now that we loaded all data,
# start processing
chunk.matchPhotos( accuracy=PhotoScan.HighAccuracy, generic_preselection=True, reference_preselection=False )
#chunk.alignCameras()
#chunk.optimizeCameras()
#chunk.buildDepthMaps( quality = PhotoScan.LowQuality, filter = PhotoScan.AggressiveFiltering )
#chunk.buildDenseCloud( point_colors = True )

# now that we're done processing,
# store the results
doc.save( out_path )

However, this results in an image size mismatch error.
Am I approaching this in the correct way?
What does the image size mismatch mean?
All my images have the same size.

Any help is much appreciated.

2

Python and Java API / Importing a transformation matrix

« on: July 02, 2018, 12:35:45 PM »

Hi,

Currently I'm trying to generate point clouds for a dataset of about 500 images of which I have the corresponding camera parameters.
I'd like to use the Python interface to load my camera parameters as setting them by hand wouldn't be scalable.
As I understand I could import the camera intrinsics (lens parameters) into a sensor's user_calib field,
and I could import the camera extrinsics into a camera's transform field. 
However, I have my camera parameters available as 4x4 projection matrices that combine both the extrinsics and intrinsics per camera.
Is there a way to import the projection matrix directly, without loading extrinsics and intrinsics separately?