The analysis is in the analysis.pdf file
You can also run analysis.ipynb, it should take several seconds if you haven't deleted anything from the repo.
TLDR:
sudo apt-get install libopencv-dev python3-opencv
./mkrun.sh [options] # build and execute with default configurations
./visualize.py [ -f pointclod-file.xyz] # visualize a pointcloud
To only make the project:
./make.sh
To only run the project with default configurations (stereo extraction):
./run_stereo.sh [options]
or
./build/OpenCV_stereo [options]
To describe parameters:
./build/OpenCV_stereo --help
To visualize points from a specific file only:
./visualize_cloud.py [-f INPUT]
Parameters for ./build/OpenCV_stereo are also set in params.cfg file, that is also explained in ./build/OpenCV_stereo --help message.
Parameters passed from the cli have more priority than the ones from the config file.
Check out --help option of the script to see the available metrics and usage.
For the naive method, there is an old script to compare performance for different w_size:
./analysis/compare_disparities.py
It will output the list of values for each metric. In the list, values are compared for different window sizes (3, 5, 7, 9), you can see my result saved in output/metrics.txt.
Input images:
Result of Naive Approach (on the left) and Dynamic Programming Approach (on the right) for Stereo Vision:
Resulting Pointcloud (filtered):
