I came across this and I'm just intrigued as to how it actually dose the tracking. The FAQ says it uses computer vision to do the tracking but it also requires the user to have a tracking device on them. So I'm guessing that the tracking device is sending out a signal to give the drone basic orientation but what sort of signal?

 

it says sensors GPS, Microphone, so it will use those.

 

The thing is that it keeps the camera pointed at the user and would seem to be at a more or less fixed distance from the user. GPS is often only accurate to the order of 10m or so and how would a microphone help? I assume the microphone is for audio recording as part of the video.

I can see that computer vision is probably used to set the distance to the subject but if the CV lock is lost it needs some way to re-acquire it. The question of "how do I get an alert if my laptop/dog/child, etc. exceeds a certain distance from me" seems to crop up on here reasonably frequently and may be this system has a solution to that question.

 

The microphone is on the users wristband, not on the camera/drone. So I am guessing
- the drone emits a unique ultrasonic pattern (or even a megaphone to direct the noise of the motors)
- the wristband detects the sound,
- the wristband sends a control instruction back to the drone to rotate/move. If not helpful, it turns the other way to rotate and move until intensity of signal is maximized again.
- Repeat.

 

Sounds plausible, I guess it has to be sound or radio because something like infra-red could become obscured by clothing, etc.

 

The microphone is on the users wristband, not on the camera/drone. So I am guessing
- the drone emits a unique ultrasonic pattern
- the wristband detects the sound,
- the wristband sends a control instruction back to the drone to rotate/move. If not helpful, it turns the other way to rotate and move until intensity of signal is maximized again.
- Repeat.

Sounds about right, this is the way I would do it. To a reply earlier, you can get GPS's with better accuracy then 10m... some down to 5m or 1m. But your tradeoff is "locking time"

 

Ditto. Ten meters would be unacceptable for many GPS applications. I have a <$20 device that's good to ~1m depending on conditions.

 

Well you can get GPS down to mm accuracy but it takes hours. For consumer grade devices as you say "depends on conditions", would you want your video to be off by 5m becuase of "conditions"?

 

I would use the video camera signal and run some sort of algorithm similar to face, smile, eye detection used in Digital cameras these days to detect a human body shape and follow that. However, a simpler method would be to place some kind of small reflective material on something visible the person would carry for the camera to track that.

 

I think it does that but first it needs to know which direction to point before it can do face recognition and it ideally pick out one face in a group.