> FaceAPI, free download for non-commercial
|
|
|
francoistarlier |
|
Class : Habitué
Posts : 121
Registered on : 16/08/2007
Off line Www
|
what do you think of that, since free-track in non-commercial right ? (except donation :()
http://www.seeingmachines.com/faceAPI.html |
|
tristan68 |
|
Class : Electronicien en chef
Posts : 651
Registered on : 10/07/2007
Off line
|
WOW , incredible 
but the drawback of this solution is (IMO) computing time to analyse each frame and detect the face.
The goal when creating a tracking system is to detect the target position ACCURATELY with a MINIMAL computing needs....
Tracking high contrast dots requires far less computing power than analysing a face to find specific points, thats a fact.
Furthermore, dots tracking is easy to setup with only minimal hardware, so its the preferred choice for a community driven headtracker (at now....).
But this tracking method is perhaps the future of headtracking as computing power will increase.... Who knows ??? |
|
cadcoke4 |
|
Class : Apprenti
Posts : 49
Registered on : 13/08/2008
Off line
|
I am curious, with the advent of multi-core processors, if the CPU burden is no longer an issue for face tracking. Since most programs aren't written to use multi cores, they would be sitting idle anyway. I imagine a 2nd core is used for things like virus scanning, and other comparable background tasks. But, a Quad-core is certainly wasting some potential capabilities.
I am still waiting for an opportunity to see if head tracking really makes a difference in my work with 3D mechanical design software. The technology has been cheap for a long time, but has not taken off. The current Samsung and Mitsubishi projection TV's have the ability to do stereoscopic with LCD shutter glasses. But, you still can't go to the store an see it in action.
Perhaps i t is like speech recognition. All the magazines from a few decades ago touted its potential benefit, and believed that the only thing preventing it from taking hold was processor speed. This technology is now mature, but is rarely used. It just didn't prove to be that beneficial, and now mostly serves to annoy us with the automated phone operator. I think the only place that voice recognition really is appreciated is for cell phones in your car.
Until I see any sort of head tracking prove its usefulness, I will try to hold back my enthusiasm... but it keeps tantalizing me.
Joe Dunfee
|
|
francoistarlier |
|
Class : Habitué
Posts : 121
Registered on : 16/08/2007
Off line Www
|
tristan68 @ 20/09/2008 - 08h42 a dit: WOW , incredible
but the drawback of this solution is (IMO) computing time to analyze each frame and detect the face.
The goal when creating a tracking system is to detect the target position ACCURATELY with a MINIMAL computing needs....
Tracking high contrast dots requires far less computing power than analysing a face to find specific points, thats a fact.
Furthermore, dots tracking is easy to setup with only minimal hardware, so its the preferred choice for a community driven headtracker (at now....).
But this tracking method is perhaps the future of headtracking as computing power will increase.... Who knows ???
well you are kind of right, but todays head tracking are less cpu consuming than it as been before, especially this one. As Video games company working with this technology already.
blob tracking have CPU advantage but its a big pain in the ass to setup as you know lighting, devices, accuracy... (if one blob is lost the all tracking is mess up. with head tracking, even if you see a quarter of the face on screen the tracking still going on.)
the biggest time consuming is Head recognition, but as soon as it's done, the tracking engine do the job and doesn't use that much cpu.
Just saying, this is a quite good API, if you wonna be the first to have this kind of features you better work on it asap  and I'm not saying to totally replaced the blob tracking, but maybe proposing both at first. Because in about a year or maximum 2 years. Blob tracking is going to be old story and you should think about Free-track future soon enought
(I can't tell you more of what I know and why am I so certain myself about this kind of technology in video games, but let's say I'm working for a video game company ^^).
good luck anyway
|
|
dzid_ |
|
Class : Apprenti
Posts : 2
Registered on : 17/10/2008
Off line
|
It would so great to see FaceAPI in Free-track.
|
|
Quarrion |
|
Class : Beta Tester
Posts : 79
Registered on : 09/10/2008
Off line
|
they probably wouldn't let freetrack have it for free because then other programs could effectively get faceapi for free through freetrack
|
|
Quarrion |
|
Class : Beta Tester
Posts : 79
Registered on : 09/10/2008
Off line
|
The demo and api have been released for public download! (100MB demo)
The demonstration videos were very impressive and probably increased my expectations too much. Even though my webcam is ir modded (very sensitive to light) and using infrared lighting to keep my face evenly lit, it still has trouble finding my face and tracking it. If it loses tracking it can sometimes take a while to get it back again (not very practical in a game). It uses a LOT more cpu than advertised (not sure why) and the all-important yaw/pitch measurements are pretty shaky and have limited range so it looks like it has a while to go before being a serious contender.
Still, compared with other face trackers this one is amazing and it will only get better with more powerful processors.
|
|
cadcoke4 |
|
Class : Apprenti
Posts : 49
Registered on : 13/08/2008
Off line
|
There is a camera called the "Z-Cam" which can extract the depth of an object. [ www.3dvsystems.com ]It can use this information to remove any background objects beyond a certain distance. This should make head tracking much easier for a computer.
The Z-cam works by emitting a pulse of IR light, and then controlling when its light sensor shuts off very precisely. For example, it emits a pulse of light that bounces off a person's face and is focused on the imaging sensor. But, the sensor is shut down before any light from further away has time to bounce back.
If I recall correctly, it was supposed to be out this year and be marketed to gamers. But, their web site has not had any updates since last year. The last "news update" is Dec, 2007. They had working products last year, so I am bit surprised they don't have it out yet.
Joe Dunfee |
|
Quarrion |
|
Class : Beta Tester
Posts : 79
Registered on : 09/10/2008
Off line
|
zcam ~ $100USD, 60fps but would still require a faceapi-like tracking software which would chew through the cpu,
better-off using a cheap webcam and faceapi imho
|
|
Eclipse |
|
Class : Apprenti
Posts : 2
Registered on : 18/11/2008
Off line
|
http://www.seeingmachines.com/licensing.html
"Non-commercial means you are not being compensated in any form for the application you develop using Seeing Machines faceAPI."
"If you release, transmit, or make available any application you have written using the Non-Commercial license, you must also release the source-code for your application, as well as the faceAPI non-commercial license in its unmodified form."
Since FT is already a free download, and the source is available, FT qualifies to be able to use faceAPI.
It would be nice to see this technology out and "running rampant".
|
|
ogoid |
|
Class : Apprenti
Posts : 2
Registered on : 05/02/2008
Off line
|
And how about a simpler detection algorithm? We don't need a full-blown face tracking, just one that can keep tracking of at least three points on the image.
OpenCV has an well developed api that could be easily integrated. In particular check the 'lkdemo' sample using a modified Lucas-Kanade optical flow algorithm.
I think that getting rid of the necessary tracking hardware really would be a great new 'feature'.
|
|
Quarrion |
|
Class : Beta Tester
Posts : 79
Registered on : 09/10/2008
Off line
|
The biggest roadblock to FaceAPI is the size of it: 100MB, which needs to be distributed with the program. So for FreeTrack that would total to a bit over 130MB. Hosting cost for this would be a drain on a free project.
Edit:
@ogoid
I wouldn't put much faith in lkdemo judging by this video:
http://au.youtube.com/watch?v=1ojq2E5A-Zg
Edited by Quarrion on 24/11/2008 at 08h17. |
|
ogoid |
|
Class : Apprenti
Posts : 2
Registered on : 05/02/2008
Off line
|
Look at these:
http://au.youtube.com/watch?v=hm-FWnr5UZU
http://au.youtube.com/watch?v=GDINqY9kekU
Sure some work would be needed to get a way to distinguish between the head rotation from its dislocation. One way that I thought was to use the camshift algorithm to detect the read, and them create some (random?) points inside it. When the points move to one side or another, keeping the same bounding box center, we have an indication of rotation. I have no clue if this can distinguish precisely the rotation from the head's dislocation (and the amount of them), but by this way I don't think that we would have any problem with the "slippage" of the points.
Besides that (working it or not), I think that FreeTrack could be improved by implementing other image filters. The current method based on the light intensity needs some light source (and energy). Instead, the blob tracking mechanism could filter the image based on some color, or by something like the camshift algorithm. By this way the tracking hardware could be anything with a specific color (instead of leds).
Edited by ogoid on 30/11/2008 at 22h22. |
|
cadcoke4 |
|
Class : Apprenti
Posts : 49
Registered on : 13/08/2008
Off line
|
FreeTrack does track by color... at least if you use a filter in front of your camera. A good quality "notch filter" can be very specific about the color. Of course, the IR light source (or reflector) is the best, since most environments will have little of that to compete with the tracking points.
Joe Dunfee
|
|
Valen2267 |
|
Class : Apprenti
Posts : 4
Registered on : 01/01/2009
Off line
|
Hi, this is my first post....
I downloaded the FaceAPI program, and was trying to download the FreeTrack source but it seems to be down for the moment...
I'm trying to get FaceAPI to output its information as 3 points of IR light (the two eyes and the lips perhaps?) so that FreeTrack can interpret the data. Anyone think this could work? I don't really know what I'm doing since I'm not a programmer but when I get the FreeTrack download to work I'm gonna look through the source code.
|