Darknet Highlights & Post-Processing - Maui63

The following library processes Maui63's output data by running their darknet model on media files and combining the output with UAV logs.

---

Installation

To install the package, simply run:

pip install git+https://github.com/Christophe-Foyer/maui63_postprocessing.git

Or if you cloned the repo, from the root directory, run:

pip install .

Note: The example and test files assume the darknet files are in the repository's parent directory.

TODO

• Upload video clips to rvision (waiting for API)
• Improve upload performance (asyncio or threads)
• Add option for object data format as independent columns for each object (instead of lists inside columns)
• Fix padding for start/end clips when creating highlights.
• Push data to R/Vision
• Speed up opencv processing? (currently ~3 fps on a GTX 960m, might just be my GPU)

Usage

The media file(s) are automatically tagged and objects added to a pandas dataframe.

In the case of a video input, if a directory is specified as an output, highlights will also be generated (see examples for highlighter args).

To create a data processing instance and run it:

from maui63_postprocessing import Maui63DataProcessor 

processor = Maui63DataProcessor(
        uav_logs,       # CSV file for UAV data logs
        media_file,     # video/image file or image folder
        data_file,      # darknet .data file
        config_file,    # darknet .cfg file
        weights_file,   # darknet .weights file
        names_file,     # darknet .names file
        output_path     # ouput file/directory
        )
        
# Run the process routine
processor.process()

To export the dataframe to a csv file:

processor.export_csv(csv_output_path)

To export the data to rvision:

# export data with a minimum spacing of 30s between frames
processor.export_rvision(post_url, min_spacing=30)

The format is "https://be.uat.rvision.rush.co.nz/api/v1/alpr/camera/<camera>/analyse-image/?token=<camera_token>"