kevinmaltezcavalheiro / tracking-droplets

This directory is dedicated to the implementation of algorithms aimed at detecting droplets in microfluidic images.

Droplet Detection Algorithm

This project consists of two separate codes. The first one is an algorithm I developed based on conventional image analysis. The second one uses the Detectron2 deep learning algorithm from Meta group https://ai.meta.com/tools/detectron2/.

Using the first algorithm

Here are the main steps for installation and usage.

1. Create a PyCharm environment and verify that the virtual environment folder (venv) has been created.
2. Clone the project from GitLab using either SSH or HTTPS. For example:

cd /path/to/your/python/project/and/venv/directory/
git clone [email protected]:K.Maltez/tracking_droplets.git

3. Open the project in PyCharm. The relevant files for this model are training_svc_blackhat.py, tracking_droplet_blackhat.py, and the svc directory.
4. Install necessary libraries by running the following command in the PyCharm terminal (make sure to install them in the virtual environment):

pip install matplotlib numpy opencv-python time pandas scikit-image scipy scikit-learn joblib trackpy mplcursors

To use the code, follow these steps:

1. Navigate to the files dictionary in tracking_droplet_blackhat.py. Change the path to your video file of choice. You can save multiple paths as "video1", "video2", etc., to avoid retyping it each time you change videos.
2. In the main section, specify your video with video = "video7" (for example). This corresponds to the previously defined path.
3. Still at the beginning of main, adjust the main parameters such as pixel size and frames per second (fps) according to your video. Also choose which images to load through the video_sampling, start_frame, and end_frame variables. Note that changing the sampling rate will also affect the fps.
4. Run the code! The detected drops should appear in green.

For more information about how the code works and how to adapt it to your needs, see algorithm1.md. For the use of detectron2, see algorithm2.md.

As this is a first version, feel free to provide feedback regarding optimization or potential bugs. I will do my best to correct them over time.

Here is an example of droplets detections based on the first algorithm

With pytrack, it is also possible to track droplets frame by frame while watching the video :

Moreover, you can plot the histogram of different velocities measured:

And here another example is provided by using detectron2 (note that I forgot to change 'balloon' to 'droplet' following a tutorial on using detectron2 https://gilberttanner.com/blog/detectron-2-object-detection-with-pytorch/)