This project runs mainly on Python. It uses the following packages:
- OpenCV for image processing
- NumPy for numerical processing
- Matplotlib for plotting
- PySide2 for the graphical user interface
Note: PySide2 (aka Qt for Python) is the ofifcial Python wrapper for Qt. The GUI runs on Qt.
The project has been tested on Python 3 and may not work on Python 2.
These packages can be installed e.g. using pip3. On Ubuntu:
sudo apt install python3-pip
sudo pip3 install opencv-python numpy matplotlib PySide2
Raspberry Pi runs on Raspbian which is based on Debian. Installation is not as straightfoward as some packages are not easily available. The required packages have been compiled in the file install.sh:
chmod +x install.sh
sudo ./install.sh
Then, in the Raspberry Pi's ~/.bashrc file, add the following line
export LD_PRELOAD=/usr/lib/arm-linux-gnueabihf/libatomic.so.1
and run
source ~/.bashrc
to reload bash. This has been tested on Raspbian GNU/Linux 10 (buster).
The silver connectors should face the HDMI port. First enable the camera using
sudo raspi-config
To test whether the camera is correctly connected one can run vcgencmd get_camera.
To take a test image one can run raspistill -o test.jpg.
When connecting over e.g. Ubuntu use X forwarding using the -X flag.
python3 main.py
A window such as the following will appear:
Choose the desired image by clicking browse (followed by load). Choose the desired parameters on the right and click the 'Find contours' button. The image will be updated and the program will display the number of bands detected in the ladder region. The user needs to manually key in the base pair values for the ladder region, ensuring that the number of values keyed in matches the number of bands detected. Then click the 'Calibrate and fit data' button and the results will be updated in the table below.
Sample output after performing all the above steps:
The code first performs colour quantisation followed by binary thresholding. It then finds contours and performs PCA analysis to obtain the centres of each contour region. It assumes that the reference ladder is on the left. After collecting the calibration data it tries to obtain a line of best fit on a semilog graph.
OpenCV: