Tools for the Generation and Visualization of Large-scale Three-dimensional Reconstructions from Image Data.
structured requires:
- cmake (>=2.8)
- openscenegraph (>=3.0)
- OpenCV (>=2.2)
- OpenGL
- TCL/TK (8.4)
- vips
- gdal (including python bindings and binaries)
- csh
For installing on Ubuntu the following apt-get install lines should pull in the requirements:
sudo apt-get install cmake libcv-dev libopenscenegraph-dev gdal-bin
sudo apt-get install libgdal1-dev python-gdal libvips-dev libopencv-dev
sudo apt-get install libhighgui-dev libcvaux-dev libopencv-gpu-dev
sudo apt-get installlibopencv-gpu2.3 tcl8.4-dev tk8.4-dev csh imagemagick
sudo apt-get install cmake libcv-dev libopencv-dev libopenscenegraph-dev
sudo apt-get install gdal-bin libgdal1-dev python-gdal libvips-dev
sudo apt-get install libhighgui-dev libcvaux-dev tcl8.4-dev
sudo apt-get install tk8.4-dev csh imagemagick
Binary is not installed and must be called using the full path of the source dir.
For example:
/home/$USER/git/structured/threadedStereo .
Once data folder is setup as below.
| Filename | Desc |
|---|---|
| mesh.cfg | Reconstruction config file. A default file is in the structured directory |
| img | A link to the converted images |
| stereo.calib | The stereo calibration file (this should be in the renav directory) |
| localiser.cfg | The SLAM config file (this should be in the renav directory) |
| stereo_pose_est.data | A link to the file stereo_pose_est.data produced by seabed_slam |
The meshes can be viewed using the program benthicQT also avalaible on github.
The AUV estimates its position in a local navigation reference frame (not for example in degrees of latitude/longitude). Positions in the navigation estimate files are measured relative to the origin of the navigation frame.
The orientation of the local navigation frame is defined as follows:
-
Positive X is true (not magnetic) North
-
Positive Y is East
-
Positive Z is down
The origin of the local navigation frame is defined by latitude and longitude coordinates.
Each line describes the pose of the stereo-vision system relative to the navigation frame at the time a pair of stereo images were acquired.
The stereo-rig reference frame is defined to be coincident with the left camera..
The orientation of the stereo-rig frame is defined as follows:
-
Positive X is aligned with the X-axis of an image (towards the right of an image)
-
Positive Y is aligned with the Y-axis of an image (towards the bottom of an image)
-
Positive Z is the direction in which the camera is looking
On each line of the file are 13 items:
| Position | Desc | Type |
|---|---|---|
| 1 | Pose identifier | integer value |
| 2 | Timestamp | in seconds |
| 3 | X position (North) | in meters, relative to nav frame |
| 4 | Y position (East) | in meters, relative to nav frame |
| 5 | Z position (Down) | in meters, relative to nav frame |
| 6 | X-axis Euler angle (Roll) | in radians, relative to nav frame |
| 7 | Y-axis Euler angle (Pitch) | in radians, relative to nav frame |
| 8 | Z-axis Euler angle (Yaw/Heading) | in radians, relative to nav frame |
| 9 | Left image name | local filename |
| 10 | Right image name | local filename |
| 11 | Vehicle altitude | in meters |
| 12 | Approximate bounding image radius | in meters |
| 13 | Likely trajectory cross-over point | 1 for true, 0 for false |
Data items 12 and 13 are used within our 3D mesh building software, and can safely be ignored in other applications.
-
First Line: number of cameras (1 number)
-
For each camera a line with camera parameters
- image size (2): [x y]
- intrinsic parameter matrix (9): [ fx 0 cx; 0 fy cy; 0 0 1 ]
- distortion coefficients, 2 radial then 2 tangential (4): [k1 k2 p1 p2]
- rotation matrix (9)
- translation vector in meters (3)
-
Rectified image quadrangle for camera1 (8)
-
Rectified image quadrangle for camera2 (8)
-
Transformation Coefficients for camera1 (9)
-
Transformation Coefficients for camera2 (9)
Notes: for a stereo config, the number of cameras is two. for stefs calib, the left camera is camera1, the right is camera2
// Number of cameras
2
// Camera 1 - Image size
1360.0000000000 1024.0000000000
// Camera 1 - Instrinsic parameters [ fx 0 cx; 0 fy cy; 0 0 1 ]
1742.4833984375 0.0000000000 702.2943725586
0.0000000000 1744.5247802734 505.3277282715
0.0000000000 0.0000000000 1.0000000000
// Camera 1 - Distortion parameters [ k1 k2 p1 p2 ]
0.1640239209 0.6399505734 0.0019153117 0.0084038246
// Camera 1 - Rotation matrix
-0.9471259117 -0.3171964586 -0.0483626388
-0.3193612993 0.9173640609 0.2375952154
-0.0309982132 0.2404777408 -0.9701595306
// Camera 1 - Translation vector
0.1716919392 -0.0959866196 1.1181910038
// Camera 2 - Image Size
1360.0000000000 1024.0000000000
// Camera 2 - Instrinsic parameters [ fx 0 cx; 0 fy cy; 0 0 1 ]
1739.3259277344 0.0000000000 678.7371215820
0.0000000000 1740.5989990234 524.8011474609
0.0000000000 0.0000000000 1.0000000000
// Camera 2 - Distortion parameters [ k1 k2 p1 p2 ]
0.1479872018 0.7572427392 -0.0015344607 -0.0037570531
// Camera 2 - Rotation matrix
-0.9497462511 -0.3011870682 -0.0852552578
-0.3128538132 0.9222519994 0.2270985991
0.0102276700 0.2423584610 -0.9701328874
// Camera 2 - Translation vector
0.1418329179 -0.0863670856 1.1126987934
// Camera 1 - coordinates of destination quadrangle after epipolar geometry rectification
-63.7492790222 -17.5878181458 1363.6937255859 0.0580182485 1360.0000000000 983.3759155273 -63.3137931824 990.2045288086
// Camera 2 - coordinates of destination quadrangle after epipolar geometry rectification
-3.4339661598 35.0206069946 1416.6680908203 25.5735092163 1423.8615722656 1022.9803466797 0.0000000000 1023.4026489258
// Camera 1 - Transformation coefficients
1.0745683575 0.0002419059 -63.7492790280 0.0129786324 0.9870401045 -17.5878181530 0.0000183164 0.0000028963 1.0000000000
// Camera 2 - Transformation coefficients
1.0347662652 0.0033534826 -3.4339661738 -0.0071165600 0.9626237039 35.0206069774 -0.0000066539 -0.0000025338 1.0000000000
The stereo-rig Z-axis Euler angle (shown as φ in the figure below) is the angle between the X-axis of the navigation frame (north) and the X-axis of the camera or image. The image rotation angle (shown as α) is the angle between East and the camera/image X-axis.
The image rotation angle can therefore be obtained from the Z-axis Euler angle by subtracting 90 degrees:
α = φ - 90°
In the figure below, the stereo-rig's Z-axis Euler angle is 135°, and the image rotation angle is 45°.
- Align the CAD model to a the same frame as the stereo_pose_est file.
- Ensure that the face normals are pointed towards the exterior of the surface.
- Remesh the surface to have ~7cm vertex resolution. For instance, use the midpoint resampling method in meshlab.
- Pass the '--manual_total_ply ' option to threadedStereo to use this instead of vrip-ing the individual stereo-derived surfaces