The hardware in Kinect 2.0 looks to be amazing where is the software to show it off?
- Thread starter onQ
- Start date
D
Deleted member 11852
Guest
More pixels don't always translate to better images.
http://www.clarkvision.com/articles/does.pixel.size.matter/
I know this already I been searching trying to find out what sensor is in the Kinect & PS4 Camera the last few days.
So far I have narrowed the PS4 Camera sensor down to being OV9714 it has a 1/4in image sensor ( if that's the sensor) haven't been able to match the kinect sensor with anything yet but from what I have seen the Kinect camera has really good image quality.
LightHeaven
Regular
Lola, at the demo. I wonder what he will be using that hand on XD
LightHeaven
Regular
The resolution might not seem like a huge increase compared to Kinect1, but the depth map is of much higher res since they switched to tof instead of structured light.
Not all demos show, but the face almost seems recognizable even by looking directly at the depth buffer, and the active IR image should be even better for it.
AMD is making Gesture control & facial recognition available for people with compatible AMD APU's that will work with your normal webcams even in low light.
http://www.amd.com/us/consumer/software/Pages/gesture-control.aspx
See what I mean by Microsoft not really showing what the hardware that went into Kinect 2.0 is enabling beyond the fact that they have more processing power in the Xbox One now compared to Xbox 360.
I'm not saying that the IR camera isn't good enough to do facial recognition I just think that the HD 1080P camera is the 1st line of defense compared to the 512 x 424 IR mode when they are doing the facial recognition with the IR camera stepping in to help.
http://www.amd.com/us/consumer/software/Pages/gesture-control.aspx
See what I mean by Microsoft not really showing what the hardware that went into Kinect 2.0 is enabling beyond the fact that they have more processing power in the Xbox One now compared to Xbox 360.
The resolution might not seem like a huge increase compared to Kinect1, but the depth map is of much higher res since they switched to tof instead of structured light.
Not all demos show, but the face almost seems recognizable even by looking directly at the depth buffer, and the active IR image should be even better for it.
I'm not saying that the IR camera isn't good enough to do facial recognition I just think that the HD 1080P camera is the 1st line of defense compared to the 512 x 424 IR mode when they are doing the facial recognition with the IR camera stepping in to help.
I skimmed what you posted but it's all talk of pc, so presumably the webcam is just a foot or so from the users face which implies the users face taking up most of the fov. Have they shown all that working from 10 feet away when the users face and gestures occupy single digit percentage of what the sensor has captured? Also right in the text you quoted is:
Seems like limited usefulness in the real world no?
He showed it being used at long range & close range.
Did you ignore what was being said about the low light filter?
"Plus, get optimized accuracy in low-light conditions that puts other systems to shame"
Watch the video.
Now imagine the demo guy was black. Would the hand tracking really work against his dark suit? I also doubt their claims about it working in low light. If they have a high enough framerate+resolution sensor then it can as they can solve noise issues (but they're just using whatever webcam the user has), but every claim of low light performance in a camera device has been bunk for the past ten years IMO. 50% accuracy at low light may be good in computing terms but it's lousy for end users and would generally be considered broken. Their example is 12" from the webcam giving 95% accuracy which isn't comparable to a console experience.
If someone wants to run the PC apps and report on how well they do, and prove that camera vision can achieve the claims in a wide variety of test cases, then I'll be happy to be proven wrong. Until then, AMD will be showing the best of their offering and the real results will be somewhat less. How much less, and whether it really can go toe-to-toe with K2 in terms of accuracy and speed, is the key question.
I'm black & I made a demo of the PlayStation Eye tracking my hands in the dark only using the light from my laptop screen.
If you're using something like this you're going to be in front of a TV or Monitor the key would be to set the camera in a mode where it would only detect the light reflecting off your skin from the TV or other light source.
On a side note: The guy talked about also using 2 cameras which would be like the PS4 Camera so if this become the norm & laptops & tablets start shipping with stereo cameras the hardware & software would soon go beyond Kinect 2.0 which is already limited to 512 x 424 30FPS depth sensing.
Last edited by a moderator:
I did, that video has lots of light compared to the situations that kinect has to work in where some people game with very little light, or back light, or lots of other situations. They setup a very controlled single situation and put it on video, big deal. Also his "gesture" is always the same, there's no way to know if it's detecting what his fingers are actually doing or if it's just detecting some kind of change in the hand area and taking that as a click. Heck even the depth demonstation was inaccurate, it was moving in Space Harrier type steps like they were approximating things just by his hand size. They didn't even show if it can track multiple people and gestures. I don't see how this video demonstrates this as being even remotely a challenge to what kinect can do. What they showed seems more applicable to tablets or phone use, not use in game rooms.
It doesn't, and I never said it did. What's shown isn't at all comparable IMO and I doubt it will be. However, I only said, "Watch the vid," because you assumed it's working at laptop-type distances. If you watch the vid, you wouldn't say that - it's working with its moderate tracking at full TV distance with at least enough accuracy to get some finger-pressing UI interaction.
Ok, I just didn't consider what they showed to be working from a tv distance. Given the thread title I was expecting to see kinect type functionality in more challenging environments with multiple people from a tv distance and didn't, and hence was left confused. What they are showing I think other people have offered in the past, maybe even some Samsung tv's have had that built in I believe. It's much more limited in scope, but ultimately has little to do with what kinect can do.
