Mr Latte said:
Oh my lord, you just can't help yourself.. so I may as well continue since you refuse to take this to e-mail, but want to keep the ad-hominem responses going.
I'll just keep re-iterating until the penny drops, discuss it civilly and stop the insults, I've kept it civil in the last few posts, and aim to do so, we can just address the points in a calm manner from now on.
So.. For the ten millionth time, and I'm happy to keep cutting and pasting this for all future replies.
The Original Video shows a robust, highly accurate head tracking system that just happens to use a Wii-mote, however an almost identical system is used commerically (TrackIR), an accepted top quality head tracking system.
You saw the video and just saw 'head tracking', I saw the video and saw a demonstration of a proper head tracking system one that is robust and accurate in all conditions and situations.
You then said that the PS-EYE could do what that system does
I pointed out that as it was, the PS-EYE couldn't match that system without some addition to the system,
NOT THAT IT COULDN'T DO A ROUGH APPROXIMATION OF IT
I've said nothing other then highlighted that a normal camera and facial recognition does not offer the same robustness or accuracy as the proper systems, which I've done time and time and time again.
Funnily I've been using head-tracking in 3D for 2 years with my web-cams, so have always been fully aware of such fundamental camera based capabilities, I can't believe you think I need reminding of this by posting so many web-cam based head tracking videos which just show the same non optimal approximation head tracking algorithms based on facial recognition.
You've totally mistaken this as me saying that facial only head tracking can't be done,
THIS IS NOT WHAT I AM SAYING/HAVE SAID OR OTHERWISE, I don't care if you infer this from what I say, I've repeatedly told you this is not what I have said/meant from the beginning, and tried many many times to show you why the two systems are different, but you failed to comprehend this even when every point is backed up with evidence or proof.
You take things so literally all the time, and even when Sony's R&D guys say that Move was invented to overcome poor lighting/detection robustness that eye-toy suffered from. You then pick on the literal fact the guy said eye-toy and that Move is a PS-EYE, which is irrelevant, it's still a camera, it's still got limited dynamic range as a light sensor, and clearly this is still the case as they've lit the Move wands, as even though the PS-EYE is better performing then eye-toy, clearly it's still not robust enough without an illuminated target, which anyone with an ounce of intelligence and common sense (as well as Sony saying that's why it's illuminated) can see it's to allow it to work in a vast range of lighting conditions (static and dynamic).
Once again, I'll re-iterate why the two systems are different, and if you keep insulting and retorting in such a coarse manner, I'll just keep on doing so.
In simple terms, comparing the head tracking ability of PS-EYE to the Wii-mote method or commercial head tracking like TrackIR
1.Does the Wii Mote/Track IR methods work in all light conditions - Yes
Why? They use Infra-Red for detection
2. Does the PS-EYE work in all light conditions - No
Why? as with all normal camera's poor light reduces the information available to the sensor
Proof - Anton Mikhailov (The Sony guy who developed the PS-EYE) said
Quote:
"One big issue with EyeToy we always tried to tackle was lighting. If you have low-light conditions, you can't see the user and you can't track him very well. That's why the spheres are illuminated: you can work in pitch-black conditions. Second thing: it's robust. It goes back to precision: if the interface isn't precise, the user starts to blame the interface and we don't want that.
Note that no IR or low light technology other then the illumination of the wands has been given, which really is because it is just a camera, and that's all.
3. Does the Wii-Mote / Track IR systems offer a high degree of accuracy - Yes
Why - For high accuracy they both use a fixed 'target' that moves with the users head. Since the target is two or more point sources of light, of a fixed size and distance/angle apart, the camera can very accurately determine the exact centre of each target light, and knowing the geometry can work out the exact 3D position, depth/tilt/angle, etc
4. Does the PS-EYE offer the same accuracy - No
Why? The proposed head tracking scheme would use pure facial recognition, this uses facial features to determine the distance/angle of your head. The problems in terms of accuracy are that detecting the exact centres of each facial feature causes a degree of error, how wide/closed your eyes are, any changes in light that casts shadows around the features, all lead to a jittery approximiation of each facial feature that is changing due to many variables, since these facial features are quite close to each other, just a few mm of error in determining one of the facial feature target points results in a much larger error.
Evidence to back that up? - Sony has given us the exact nature of why they chose the 'wands' with spheres on top. Firstly they are illuminated, so any change in light conditions or shadows will have negligible effect, secondly, no matter what angle they are at, the diameter of the shere is a constant, the camera measures the diameter and since it knows the exact sphere size, can place it instantly in 3 dimensions, this is almost indentical in principle to the TrackIR and Wii-mote target marker principle, i.e. a fixed geometry marker to allow accurate and fast 3D positioning data.
5. Can PS-EYE do head-tracking then? Yes
Why? It's been around for ages using just webcams, PS-EYE is a fast/hi-res web-cam. The logitech HD web-cams already offer 3D Avatars, where the camera is tracking the head reasonably well, it tracks your mouth (smile and your avatar smiles, look sad, your avatar looks sad), it tracks your eye movement (wink, and your avatar winks). This type of technology has been around for some time. However, as can be seen above it is not as reliable, robust or accurate as the more dedicated solutions.
But, that shouldn't stop games like GT5 or as per the shown head tracking demo's working well enough, providing they confine the face with reasonable 3D bounds, and obviously the user's lighting scheme is sufficient, then it'll give a reasonable head tracking experience. If you want it to work reasonably well while moving all around the room, and not have sudden moments of uncertainty (the main reason this type of far field tracking is reserved for window washing type games), then it's going to not be as robust by a long chalk as the Wii-Mote/TrackIR type solutions.
6. Have I only ever said that the PS-EYE cannot offer the same accuracy and reliability as the Wii-mote method shown in the video (unless some type of target device is used as well)? Yes
I can't believe you really have shunned an open and honest attempt to resolve and further refrain from this discussion, you really are not willing to remotely see anything but your own view point.
I understand exactly what you are trying to show, that 2D Cameras can do head tracking via Facial recognition, which I totally agree with, however that's not what I originally meant or said, I have only ever been concerned with classing 2D camera facial recognition systems (.e.g. PS-EYE/web-cams) with 'proper' accurate/robust systems).
I've got the Cut and paste ready for no doubt the next tirade.
Why is it I have no issues with accepting what you 'mean' and you refuse to accept what I mean, which are clearly two different things, like you want to keep this going or something?
As soon as I realised you where taking my comments the wrong way and thinking I was saying it was 'not possible' I immediately clarified it to stop this confusion.
Me said:
Well, my friend, what I meant was the PS Eye does not work in all lighting conditions, to whit exactly why 'Move' has lights on it for tracking in all lighting conditions, hence why I said that to be ensured to function correctly, it would really require some point light sources to ensure reliable detection.. Also it has no depth sensor, other then approximating from a 2D face image, again, something that is unreliable and means 3D spacial awareness is reduced.
But still, despite clarifying immediately, you still have gone on and on and on accusing me of saying I said it was not possible at all, which clearly is rubbish, all my posts are in this thread, unaltered for you to read, I agree that one post was ambigious, and could have been taken out of context, but the above was my clarification of this, and you have ignored it, and just now gone down a dead end as far as I am concerned, I've done everything I can to keep clarifying what I meant, what more can be done?
