This edge detection system is designed to digest images captured by the SkyScan system and determine whether there is an aircraft fully in frame or not. A lack of aircraft in the image can occur if the aircraft the system was tracking when the image was taken was occluded (e.g., behind a building, tree, etc.) or too distant.
The edge detection system uses docker compose and runs two services, one to move files out of the file structure generated by the SkyScan system into a central location (the filemover
service) and the other to perform aircraft localization and detection (the ai
service).
In order to properly run the localization service, a weights file (commonly weights.pt
or localizer.pt
) must be placed in the /weights
directory, which will then be mounted inside of the container.
Both services can be launched using the command docker compose up
if run in this edge-detect
directory. The ai service will then begin taking raw images from /flash/unprocessed
(placed there by the filemover service) or another volume specified in its place in the Dockerfile included in the ai
folder. It then sorts them into plane
and noplane
folders also located, with the current configuration, under /flash
.
Additionally, the location of planes within each image is saved to JSON files under /flash/log
.