Intrusion Detection w/ Deep Learning


These are from memory… Details TBA.


  • ~400 samples, each including one 12MPixels photo + 2 seconds video (1440x1080, 25fps)
  • 4 kids climbing (mostly my son Charles and daughter Emily)
  • 3 seasons: summer, fall, winter
  • 5 different days and weather pattern, lighting, clothing, etc.
  • 10 different segments of fences with many chain-link configurations
  • Each sample taken from different angle. Distance from fence varies from 3m to 45m
  • Roughly half of images are of climbing (i.e. alarm), half have fences without climbing.
  • Changing environments, e.g. other kids, dog, cars, adults walking, running, etc.
  • Some tricky images, e.g. my son posed like he’s climbing behind the fence, but not actually touching the fence
  • Taken with my iPhone 6P
  • Total capture and pre-processing time: ~16 hours
  • No kid or animal was harmed in the making of this dataset

Internet photos

  • ~150 pictures of people climbing or jumping over fences, ~150 pictures of fences without climbing
  • total time to find suitable pictures: ~3 hours (surprisingly long!)
  • ~half of the images have EXIF tags, e.g. with GPS location, field of view, etc.