They're trying to fool the system into thinking there is NO person there: for situations such as intrusion detection where it's less WHO is there and more IS someone there.
That's not what Darknet / YOLOv2 systems are typically used for. If you just want to know if "a person is there", a cheap, reliable ultrasonic motion detector is a better starting point. You can do some image capture and recognition if you want to remove false positives (other animals, curtains or plants blowing in the wind, etc), but for those cases YOLO is not the best approach.
YOLO is intended specifically for high performance, so it can be used in streaming applications where a large number of objects need to be identified quickly for further processing.
Their attack (when successful) prevents Darknet from finding the patch bounding the protected person in the image stream. That's an effective defense against identification, and that's the case where this attack is useful.
Biting the hand that feeds IT
