Python API - Camera Sensor Parameters - Zero Resolution Error

[mvol12]

I have calibrated a set of cameras using my own custom software and stored the parameters in a json file.
I am using the python api to load my camera calibration from json files into agisoft and attempting to build a dense point cloud. The script loads the cameras in the correct positions, builds the sparse point cloud which again looks reasonable but then fails on the dense reconstruction giving a 'Zero Resolution Error'. I included a regionReset before building the builddensecloud as this error often relates to incorrect region size. I have also checked that there are enough features and matches approximately 3500 points per image and 5865 tie points.

In an attempt to debug this error i exported the camera parameters from agisoft (see below). I noticed that the camera sensor resolution is of width and height 0 despite being set. Also from the documentation, cx and cy parameters are relative to the image centre, but the sparse reconstruction only looks correct if these are set to the absolute value. This suggests that the sensor size is not being correctly initialised through the python api.

Any help with this would be appreciated.

Here is the script:

Code: [Select]

#!/usr/bin/python

import json
from pprint import pprint

import PhotoScan

camera_file = 'camera.%02d.json'
camera_image = 'image.%02d.png'
camera_start = 1
camera_stop = 20


chunk = PhotoScan.app.document.addChunk()

for camera_index in range(camera_start, camera_stop+1):

with open(camera_file%camera_index) as data_file:   
    data = json.load(data_file)


#Create Camera
camera = chunk.addCamera();

#Add Image
camera.label = "Camera %02d"%camera_index
photo_file = camera_image%camera_index
camera.open(photo_file)

#Get Rotation
matrix = PhotoScan.Matrix().Diag([1,1,1,1])
for i in range(0, 3):
for j in range(0, 3):
matrix[i,j] = float(data["camera"]["r"][i*3+j])

# Translation
matrix[0,3] = float(data["camera"]["t"][0])
matrix[1,3] = float(data["camera"]["t"][1])
matrix[2,3] = float(data["camera"]["t"][2])

# invert matrix
camera.transform = matrix.inv()

#Create Calibration
calibration    = PhotoScan.Calibration()
calibration.width = int(data["camera"]["width"])
calibration.height = int(data["camera"]["height"])
calibration.cx = float(data["camera"]["cx"])
calibration.cy = float(data["camera"]["cy"])
calibration.f = float(data["camera"]["fy"])
calibration.b1 = float(data["camera"]["fx"])-float(data["camera"]["fy"])
calibration.b2 = float(data["camera"]["skew"])
calibration.k1 = float(data["camera"]["distortion"][0])
calibration.k2 = float(data["camera"]["distortion"][1])
calibration.p1 = float(data["camera"]["distortion"][2])
calibration.p2 = float(data["camera"]["distortion"][3])
calibration.k3 = float(data["camera"]["distortion"][4])

#Create Sensor
camera.sensor = chunk.addSensor();
camera.sensor.user_calib = calibration
camera.sensor.calibration = calibration
camera.sensor.label = camera.label
camera.sensor.type = PhotoScan.Sensor.Type.Frame
camera.sensor.width = int(data["camera"]["width"])
camera.sensor.height = int(data["camera"]["height"])
camera.sensor.fixed = True


# Match Photos
chunk.matchPhotos(accuracy=PhotoScan.HighQuality, preselection=PhotoScan.NoPreselection,keypoint_limit=40000,tiepoint_limit=4000, filter_mask=False)

#Build Sparse Reconstruction
chunk.buildPoints()

# Reset Region
chunk.resetRegion()

#Build Point Cloud
chunk.buildDenseCloud(quality=PhotoScan.HighQuality)


Here is a snippet from the exported calibration

Code: [Select]

<?xml version="1.0" encoding="UTF-8"?>
<document version="1.2.0">
  <chunk>
    <sensors>
      <sensor id="0" label="Camera 01" type="frame">
        <resolution width="2448" height="2048"/>
        <property name="fixed" value="true"/>
        <calibration type="frame" class="initial">
          <resolution width="0" height="0"/>
          <fx>2259.2257678318952</fx>
          <fy>2258.0792439942561</fy>
          <cx>1244.9551486807138</cx>
          <cy>1035.0740703194135</cy>
          <k1>-0.078427925999999995</k1>
          <k2>0.15065549</k2>
          <k3>-0.12147871</k3>
          <p1>0.0013411323</p1>
          <p2>-6.7674315000000005e-05</p2>
        </calibration>
        <calibration type="frame" class="adjusted">
          <resolution width="0" height="0"/>
          <fx>2259.2257678318952</fx>
          <fy>2258.0792439942561</fy>
          <cx>1244.9551486807138</cx>
          <cy>1035.0740703194135</cy>
          <k1>-0.078427925999999995</k1>
          <k2>0.15065549</k2>
          <k3>-0.12147871</k3>
          <p1>0.0013411323</p1>
          <p2>-6.7674315000000005e-05</p2>
        </calibration>
      </sensor>


[Alexey Pasumansky]

Hello mvol12,

It seems that resolution of the photos is not defined neither for sensor, nor for the calibration (width and height values are zeros). So you need to check if this data is present in your source files and if yes, then whether it is correctly read and interpreted.

[mvol12]

Hi Alexey,

I can confirm that the values are correctly being set. I put the following line at the end of the input loop to write out the set values for calibration and sensor width and height.
But these values are not reflected in the output agisoft calibration file.

Code: [Select]

print("Camera %d:  %d %d %d %d"%(camera_index, sensor.width,sensor.height,calibration.width,calibration.height))

Here is the console output:

Code: [Select]

2017-02-07 16:58:24 Camera 1:  2448 2048 2448 2048
2017-02-07 16:58:24 Camera 2:  2448 2048 2448 2048
2017-02-07 16:58:24 Camera 3:  2448 2048 2448 2048
2017-02-07 16:58:24 Camera 4:  2448 2048 2448 2048
2017-02-07 16:58:24 Camera 5:  2448 2048 2448 2048
2017-02-07 16:58:24 Camera 6:  2448 2048 2448 2048
2017-02-07 16:58:24 Camera 7:  2448 2048 2448 2048
2017-02-07 16:58:24 Camera 8:  2448 2048 2448 2048
2017-02-07 16:58:24 Camera 9:  2448 2048 2448 2048
2017-02-07 16:58:24 Camera 10:  2448 2048 2448 2048
2017-02-07 16:58:24 Camera 11:  2448 2048 2448 2048
2017-02-07 16:58:24 Camera 12:  2448 2048 2448 2048
2017-02-07 16:58:24 Camera 13:  2448 2048 2448 2048
2017-02-07 16:58:24 Camera 14:  2448 2048 2448 2048
2017-02-07 16:58:24 Camera 15:  2448 2048 2448 2048
2017-02-07 16:58:24 Camera 16:  2448 2048 2448 2048
2017-02-07 16:58:24 Camera 17:  2448 2048 2448 2048
2017-02-07 16:58:24 Camera 18:  2448 2048 2448 2048
2017-02-07 16:58:24 Camera 19:  2448 2048 2448 2048
2017-02-07 16:58:24 Camera 20:  2448 2048 2448 2048


[Alexey Pasumansky]

Hello mvol12,

Camera calibration in XML format should contain <width> and <height> sections.

Also I suggest to define sensor.width and sensor.height before assigning .calibration and .user_calib properties.

[mvol12]

Thanks Alexey,

Changing the order has fixed my problem.
What are the rules for initialisation order in the python API?
I have had this problem in the past and i cant seem to find anything that describes this in the documentation.

Kind Regards,
Marco

[Alexey Pasumansky]

Hello Marco,

When you are assigning calibration to the sensor, the width and height parameters of PhotoScan.Calibration() are rewritten by the values from PhotoScan.Sensor() for width and height.