This work flow takes a large amount of computational power to run, so the run time will vary depending heavily on the specs of the system it is run on. The following graphic shows the process for this work flow. Each step can be run individually, however, it needs the input from the previous step in order to run properly.
Test data
Starting from the base directory the following commands should run each dataset:
-
For the non-geo-accurate reconstruction:
$ cd workspace/demo $ sh RunProcess.sh -a
-
For the geo-accurate reconstruction:
$ cd workspace/demo $ sh RunProcess.sh -ag
-
The output of the geo-accurate reconstruction upon successful completion is:
$ sh RunProcess.sh [-options] [- Prepping Data -] [- Running siftGPU -] [- Running Bundler -] [- Preparing PMVS -] [- Running CMVS -] [- Running PMVS -] [- Running GTransform -] [- Clean Up -] [- Done -]
-
The three-dimensional point clouds will be in the results directory upon completion. The *.ply files in the root of the reults directory are the non-geoaccurate point clouds. If GTransform was run, the geo-accurate models will reside in trans/models/.
-
The transformed camera projection matricies are also written out by GTransform, they will reside in trans/txt/.
-
The georegistration transform will reside in trans/transform.txt
-
The ply file format can be viewed in many different software packages, the one we recommend is Meshlab.
-
There are a number of algorithms that can be used for the meshing process, although, none are presented here. Meshlab provides a number of surface reconstruction algorithms that work quite well, such as poisson reconstruction.
RunProcess.sh
This script is used to run the work flow and contains a number of options for controlling how everything is run.
Usage
sh RunProcess.sh [-options]
Options
- -a
-
Run the whole work flow (equivalent to -sbp)
- -s
-
Run siftGPU
- -b
-
Run Bundler
- -p
-
Run CMVS/PMVS
- -g
-
Run GTransform
- -k
-
Prevent the script from running cleanup. This will keep all the output of siftGPU, Bundler, CMVS/PMVS, and GTransform.
- -d = 0
-
Set the GPU device number for siftGPU.
- -x = 2000
-
Set the maximum dimension for an image for siftGPU, this will cause siftGPU to down sample images larger than this dimension.
- -y = 8000
-
Set the maximum number of features for siftGPU. This is useful for running siftGPU at full resolution but limiting the memory usage.
- -f = 6111.11
-
Set the focal length of the camera used to take the images, currently assuming all images taken with the same camera at the same focal length. The focal length must be in pixel units (i.e. focal length in mm / pixel size in mm).
- -i = 0.0001
-
Set the focal length constraint weight for the bundler adjustment in Bundler.
- -c = 30
-
Set the max cluster size for CMVS.
- -l = 1
-
Set the level for PMVS.
- -e = 2
-
Set the cell size for PMVS.
- -t = 0.7
-
Set the threshold for PMVS.
- -w = 7
-
Set the window size for PMVS.
- -m = 3
-
Set the min image number for PMVS.
- -u = 2
-
Set the number of CPUs to use for CMVS and PMVS.
Example Usage
Running the whole work flow on a set of images with a focal length of 5000 pixels, using 6 CPUs, and with a PMVS threshold of 0.9:
$ sh RunProcess -a -f 5000 -u 6 -t 0.9
Alternatively, you can edit the default values in RunProcess.sh to change parameters.
Using your own data
For running the work flow in non-geo-accurate mode, your data can be run in the same fashion as the example for the non-geo-accurate test data. Remember to set the appropriate focal lengths by using -f focal length or altering the default value in RunProcess.sh .
For running the work flow in geo-accurate mode, each image must be accompanied by a POS file named imageFileName.pos. Each POS file must contain the X,Y, and Z positions of the camera frame. These coordinates should be in the desired Euclidean output coordinate system (i.e. UTM, NEU, etc).
