Face Recognition plug-in - Beta testing

[gearyt]

no boxes around faces ( wife walked in, now gone )

[S-F]

This is in line with my experiences. A glance won't do it. It seems that a face needs to be in very close proximity to the camera and looking right at it. I've had my Kinect looking at a doorway and I'm not recognized until I position myself right in front of the camera and look directly at it for a time in great lighting.

[spud]

This is in line with my experiences. A glance won't do it. It seems that a face needs to be in very close proximity to the camera and looking right at it. I've had my Kinect looking at a doorway and I'm not recognized until I position myself right in front of the camera and look directly at it for a time in great lighting.

All frames are resized to 320x240 before being processed, and for a face to be detected it needs to be at least 50x50. So yes the face needs to be close to the camera, but I can improve that by not resizing.
The resizing was there to avoid detecting from a high resolution image which would take more CPU, but I can make it optional.
For example the default resolution of the kinect camera is 640x480, so by not resizing we should be able to detect faces that are not so close from the camera.

To be detected the face needs to be frontal. However one feature I could add is the possibility to detect profile faces, full body, etc.. something more like Sighthound, but these alternative patterns couldn't be used to recognize people.

[S-F]

The more options the better I say!

You say that detecting profiles and so on couldn't detect people. How does the Kinect do it on the xbox? I have no first hand experience but I've heard that it can recognize people as they walk into a room.

I also have a question. How do multiple instances work? Let's say I have 5 cameras. How could I use them all?

And on a mostly unrelated note, Does anyone think that the naming "Kinect" had any influence from Joseph Kinecht the Glass Bead Game Master of the Hermann Hesse novel? I just finished reading that book again and the coincidence seems too great. But maybe I'm being too philosophical and reading too much into this and it's just a hip contraction of the word "kinetic".

[spud]

The more options the better I say!
You say that detecting profiles and so on couldn't detect people. How does the Kinect do it on the xbox? I have no first hand experience but I've heard that it can recognize people as they walk into a room.

I don't have the xbox, but I would love to heard some feedback from users who have one to know how well it works.

I also have a question. How do multiple instances work? Let's say I have 5 cameras. How could I use them all?

If you had 5 cameras, you would create and start 5 instances of the FaceRecognition plugin from the PLUG-INS > Manage page with 5 different names. You would then have one config page for each instance from which you would configure one camera and recognition settings, and you would have 5 set of devices.
The 5 plugin instances would run on the HS3 machine. I haven't really tested it but running one instance remotely should work as well but the only benefit I can see of running remotely is to balance the CPU load

And on a mostly unrelated note, Does anyone think that the naming "Kinect" had any influence from Joseph Kinecht the Glass Bead Game Master of the Hermann Hesse novel? I just finished reading that book again and the coincidence seems too great. But maybe I'm being too philosophical and reading too much into this and it's just a hip contraction of the word "kinetic".

no idea

[BIGMACK48]

I try using this url for foscam cam: http://192.168.x.xxx:80/videostream....tion=32&rate=0

Status: Recognition is currently running

I not seeing any video on your Manage Data training page.

Need help!!!!

Thank you

Greg Mack

[spud]

I try using this url for foscam cam: http://192.168.x.xxx:80/videostream....tion=32&rate=0

Status: Recognition is currently running

I not seeing any video on your Manage Data training page.

Need help!!!!

Thank you

Greg Mack

There is a bug preventing the "training data manager" to work correctly when the url contains "=".
Sorry for that, I will fix it soon.

[BIGMACK48]

Thanks

[BIGMACK48]

when will you get time to update this plugin?

I like to test it out before I buy it.

Thank you

[spud]

when will you get time to update this plugin?

I like to test it out before I buy it.

Thank you

sorry for the delay, I will have the fix done and released before the end of the week

[spud]

can you try version 3.0.0.4 available in the first post of the thread.

in this version I have fixed the bug for url containing the '=' character, and I have added the possibility to use different resolution for recognition

[nikku]

Hi Spud,
Any thoughts on making the learning utility runnable on a remote system (such as a laptop that has a camera)? My server runs in a rack and is headless, so it's a bit difficult to put a camera on it and take pictures, not sure if others would be in a similar situation. Just a thought

EDIT: Guess I didn't read enough, looks like you can use a remote camera now via snapshot URL, guess I'll try that later on, silly me!

[rhardcore]

Remote Plugin Instance

Hi, Spud. Thanks for making this. It is great, but it can be pretty taxing on the hardware in the S6. I have a server that is running as the NVR and has a bit more CPU to spare. Is it possible to run this plugin remotely and not on the main HS3 server (the HS6 in my case)? If so, that would be very helpful and makes this a very useful and feasible plugin.

Thanks!!!

[spud]

Hi, Spud. Thanks for making this. It is great, but it can be pretty taxing on the hardware in the S6. I have a server that is running as the NVR and has a bit more CPU to spare. Is it possible to run this plugin remotely and not on the main HS3 server (the HS6 in my case)? If so, that would be very helpful and makes this a very useful and feasible plugin.

Thanks!!!

yes it is possible to run this plugin remotely, but I guess it can be tricky to install it properly, I will try to come up with a tutorial for that.

[rhardcore]

Thanks, Spud. I have been muddling through it.
I have BlueIris installed on a beefy server.
I did an install of HS3 on that server.
Then, I installed the PI for facial recognition.
Then, I created and ran a batch file that runs a single command. It was something like: HS3PI_pluginname.exe -server <IP of S6 HS3 server>.

This is from memory, so it may be a little off in the syntax. I noticed immediately:

1. CPU cost of the facial recognition was offloaded to the BI server.
2. All the previous facial learning done when it was on the S6 is not accessible from the BI server.

So, it probably would have been better to just copy over the whole HS3 directories to the BI server instead. That would have moved the faces over, too.

Please let me know if you see any issues with this or know a better way.

Thanks!
Raymond