/ucsb-mealtime

Determining wait times at UCSB dining halls via computer vision

Primary LanguagePython

UCSB Meal Time

Determining wait times at UCSB dining halls via computer vision

Changelog

Next up

  • automate image collection based on dining hall open hours
    • start collecting images 5 minutes before dining hall opens
  • add stream capability to dining_hall_cams module

08/20/18

  • created dining_hall_cams module
    • save still images from dining hall camera feed
    • checks if dining hall is open at a given time
  • recorded UCSB dining hall operating hours into data/ucsb_dining_hall_hours.csv (source)

Data

UCSB has recently released a series of APIs, one of which provides feeds from cameras at Carrillo, DLG, and Ortega dining halls.

Here are examples of stills from each of the cameras retrieved by get_sample_images.py which demonstrates usage of the dining_hall_cams module.

Carrillo:

DLG:

Ortega:

Requirements

Data retrieval module

  • during dining hall open hours, automatically collect still images for training
  • get video feed

Computer vision modeling

  • image classification with classes {'no line', 'short line', 'long line'}
  • object detection to count number of people in line

Model retraining

  • automatically retrain model on a monthly (?) basis with additional images collected since last training