K'BOS! 🛬

Live High-Quality DSLR Capture Module for KBOS 22L/R Arrivals
@iot @edgeai @tensorflow-lite @opencv @debian @rpi4b @coral @canon
Automatically capture high-res photos for aircrafts on KBOS 22L/R Final Approach.

v1.0 (August 07, 2020): Workable on Debian x86 and ARMv7l, all clear on dependencies.

TODO

• Recompile tflite model for edgetpu and test speed increase
• MySQL/MongoDB: flight/aircraft/airline info for the plane captured, embedded in cyaned.co
• Integrated with project CYANED.CO: Switch between github.io and RPi server
• Flight info updates and auto matching
• (DEP) Sentry for boarder field monitoring, when an airplane is approaching, initialize DSLR
  • Use videos in raw_sentry to identify the least power necessary for sentry monitoring
  • Test live performance
  • Crop live feed and picture in cv2, helpful in boosting performance?
  • Single image and video object detection, screenshot, test correct rate
  • Live monitoring and detection (only binary info req), test FPS
  • Set init trigger params for DSLR

Stages Projected & Methodology

This project is projected to have multiple stages:

• DONE: Automatically capture high quality photos for any aircraft on approach on 22L/R
• DONE: Process photo: object detection and trimming raw inputs
• PENDING: Fetch flight data: obtain live arrival data from FlightAware.com and identify flight
• PENDING: Build a server for live feeds based on LAMP
• PENDING: Flexibility to setup in a new environment, parameterized setups

Notes

• Make sure Coral is unplugged when sudo apt install libedgetpu1-std
• Saves from DSLR gphoto2 --list-all-files; gphoto2 --get-files=XXX-XXX
• tensorflow cross-compilation
• objdetapp example
• tensorflow lite official guide
• tflite on raspian
• docker ref
• docker ref