ML-Motion-Blur

After recording a video, you may find that movements look choppy. Rerecording it at a higher framerate may not be an option, so what can you do to make it look smoother? You could apply a generic motion blur filter, but that often leads to messy images. Manually adding frames and animating them works, but that is impractical. This application aims to create an elegant solution to this problem. With machine learning, relevant features in the image can be detected for blurring. For example, the legs of someone running should be blurred, but not the bench behind them!

How it Works

We trained a machine learning model track objects that will likely be in motion, like limbs and cars. Using those detections, the application smooths motion by creating an "in between" frame. In this frame, objects are placed in the image at the midpoint where it was detected in the prior and following frame.

For a more detailed explanation of how everything works, see our report and video.

Installation

1. Install Python 3.9 or above.

2. Clone the repository. You may get errors related to a file named Model.pt. If this happens, then download it directly from Backend/main_site/Model.pt and overwrite your local version.

3. Install the required packages using your preferred package manager. For example, while in the project directory, enter:

pip install -r requirements.txt

How to run

Note, this build is intended to work for Windows, and requires modifications to be run on Linux.
In Backend/main_site/tracking.py, change from . import sort to import sort to make it work on Linux.
It has not been tested to work on Mac.

1. In the Backend folder, enter python manage.py runserver runserver in the terminal.
2. In a browser, enter localhost:8000 or http://127.0.0.1:8000/ as the url.
3. You should see an option to upload a video. Select a video to upload and wait. When it is done, it will be available to download.

A visual walkthrough of this process is found in both the video and report.

Alternatively, while in the Backend/main_site directory, you can run tracking.py with the following arguments to get the same result:

1: path the the Faster RCNN Model
2: file/path name, either to an image or video
3: Enter 0 for image, 1 for video
Ex.
python3 tracking.py Model.pt test_samples/person_standing.jpg 0

tracking.py also contains the necessary functions to use this application's features in your own project's.

Additional Credits

A core component of this application was the SORT algorithm. Their work is found in Backend/main_site/sort.py.