[Art]

Yeah...that's correct...

http://getaheadinrobotics.com/

[Korrelan]

I was kinda expecting something totally different.

What is this object... hold it closer so I can see it...

Pffft, what that thing needs a frikken huge zoom lens lol.

They are very cool though.

 

[Art]

If you're going to emulate a human, then for Pete's sake, put a camera in BOTH eyes instead of a bullet hole camera in the middle of its forehead!
just sayin...

[Freddy]

They look cool though - I took the liberty of adding a video to your post Art 

[Art]

I just have to wonder, and perhaps Korr might shed a bit of light, but is stereo vision (camera in each eye) far too difficult or costly a project to implement for robot vision? Is it a difficult process to enable both cameras to move in unison and in synchronicity with each other?

[LOCKSUIT]

interesting question to korr, goaty.

My take is I only need 1 eye, and I bet a wounded goat can grow up with 1 eye and learn still. It's as simple as: 2 eyes aren't needed. 1 eye can get you 3D.

[goaty]

interesting question to korr, goaty.
My take is I only need 1 eye, and I bet a wounded goat can grow up with 1 eye and learn still. It's as simple as: 2 eyes aren't needed. 1 eye can get you 3D.

Yo im here!

STEREO
  - cross triangulation   (if you have point 'x' on both cameras,  use algebra using the known camera offset to develop the 3d point.)

MONOCULAR
  - structure from motion   (discover with algebra the new camera position, then its triangulatable like stereo.)
  - structure from texture   (the heavier the texture, the further away it is.)

stereo generates the 3d a little easier, but its less adaptive than monocular,  because mono only takes 1 eye.

One last thing.
Because 3d generation is patchy at best,   if you have a value to correlate to, you can learn the 3d over time by training another network with the 3d.    Then to get your 3d result, you access this other trained network.

[Korrelan]

@goaty

I couldn’t have put it better my self. 

@Art

The single camera is form over function in this case; it keeps the design cheap, simple and aesthetically pleasing.  The head is not mobile and just requires a single image for face/ object detection and eye gaze calculation. It looked like they where using YOLO for both the face and object detection which has no requirement for accurate depth perception/ information.

As goaty mentioned there are several methods for gaining depth information from a single camera or indeed using another measuring technique. 

The reason I have two stereo cameras is that I’m experimenting with combining the two images into one visual cortex to create an ocular dominance map.

I also have two converging infrared lasers (Dambusters), the distance between the points roughly gives distance to the object/ surface.



Recognising an object is one thing… reaching for it is a totally different ball game.

 

[goaty]

One thing I often think to myself, is why we bother with all this rgb camera tech when you can just give yourself a Microsoft Kinect and have the 3d solved and not bother yourself with it.

Are we masochistic or what.

[Art]

interesting question to korr, goaty.

My take is I only need 1 eye, and I bet a wounded goat can grow up with 1 eye and learn still. It's as simple as: 2 eyes aren't needed. 1 eye can get you 3D.

@ Lock,
I asked the question to Korr for the record. (Thanks Korr).

One eye can NOT get 3D. It takes two eyes in order to converge together (perspective) on an object, to signal the brain it's the approximate location in our 3D space.

[LOCKSUIT]

HA HA HA...oh dear. You sounded like goaty, so I just went with 'goaty'.

Um, Art, I've seen videos where they use 1 camera, it builds 3D by seeing different angles.

[goaty]

Just dreaming a bit...
Theres something interesting about necks, and moving and zooming eyes,   that a genetic algorithm could end up using it automatically, if gathering more information from its sorroundings ended up granting it more success.   But maybe youd get a bit of an uncontrollable reaction for it if you didn't have the dampening.

[LOCKSUIT]

"uncontrollable reaction for it if you didn't have the dampening"

Zooms into cleavage and underwear of random human female woman. Building Semantic profile.

Strong AI could be something else, much deeper.

The old saying goes, it has a mind of its own!

[goaty]

Zooms into cleavage and underwear of random human female woman. Building Semantic profile.

that's funny.    Im thinking more of an insane jerking spinning motion that's still giving it high attentative scores, that you should keep out of sight of children.

[LOCKSUIT]

Detectives come around:

"We don't know how this happened, these things are not well understood. This is were things get fuzzy."