NeuroSurveillance - AI-Software for IP-Cameras

[neurosurveillance]

Hi,

I am Thorsten Hempel and developer of NeuroSurveillance, an AI-Software which detects motion, faces and recognises faces in your ip-camera stream. You can train the system individual faces, so the software knows when a familiar our unfamiliar face passes your cameras.

NeuroSurveillance contains also an alertsystem, that with notify you via sms, email or windows notficication that a certain motion has happened, faces have been detected or specific faces have been recognised (or not recognised to detect unwelcomed person). More alert options will come later.

I will add more classifiers from time to time, like license plate recognition, certain animal detection etc. So stay tuned for more

If you are interested, feel free to visit my website NeuroSurveillance
You can reach me through the contact form if you have an question or want to preorder first licenses, which will be free in alpha version.

Of couse you can also ask here in the forum, since I am passiv reader here for a longer time now.

 

[aristobrat]

Thank you for NOT building a cloud-based service. It's nice to see something that runs locally on owners hardware.

 

[bickford]

Very nice
can you explain me how the system works
You use a dedicated server to install the software ?

The Customer can install the software on a dedicated computer ?

Free licence is for 2 cameras right ?

this system is ONVIF ?

BICK

 

[neurosurveillance]

Thank you for NOT building a cloud-based service. It's nice to see something that runs locally on owners hardware.

Yes, all image processing is done on the local computer where the software is running on. Otherwise I would depend on third party cloud computing services where I can't gurantee for privacy safety.

It's not easy to realize high ressource consuming processes like face recognition in real time on end user hardware. But I can tell that the software is working on my 6 years old laptop on two streams simlutaneously with a reasonable fps rate. I am planning to optional perform image processing on the graphic card, which will improve the fps rates for many hardware systems.

Very nice
can you explain me how the system works
You use a dedicated server to install the software ?

The Customer can install the software on a dedicated computer ?

Free licence is for 2 cameras right ?

this system is ONVIF ?

BICK

The software will be installed on the users hardware (Windows only atm). It has a size of 100 mb currently. Then you can connect the software with the cameras via their ip camera addresses, where you can perform different image processing options. Everything happens in realtime, but you can also run video files or open directly connected video devices.

There will be a free version, that will be free for lifetime and support 2 cameras, right.

 

[bickford]

Thanks for your reply I have contacted you through your website

BICK

 

[BigFoot]

Hello Neurosurveillance

Would your system be capable of running along side a Blue Iris installation on the same server, or would it require an addition dedicated machine ?

 

[Armysurplus]

x2 on Bigfoots ?

 

[Armysurplus]

So I have direct line of sight of My Granddaughters school, and right now I have to pick Her out of the crowd . Could I teach the camera to detect Her as She rounds the corner then follow Her Home? If so I'm in.

 

[Q™]

It's all smoke and mirrors until you have something in your hands that actually works as claimed.

Looneyism #792: "It's called The Burger Syndrome"

 

[neurosurveillance]

Hello Neurosurveillance

Would your system be capable of running along side a Blue Iris installation on the same server, or would it require an addition dedicated machine ?

Hi BigFoot,

I don't see any problem with that.

So I have direct line of sight of My Granddaughters school, and right now I have to pick Her out of the crowd . Could I teach the camera to detect Her as She rounds the corner then follow Her Home? If so I'm in.

Yes, of course. But the classifier needs a decent face resolution for face detection and recognition. So depending on the distant between your granddaughter and the camera, the resolution should be high enough to respresent your daughters face with enough pixels to compare the face ids.

It's all smoke and mirrors until you have something in your hands that actually works as claimed.

Looneyism #792: "It's called The Burger Syndrome"

That's right. The classifier are already working perfectly, since I started the whole projekt with them. So if you have any footage you want to test, you can send it to me and I will execute it in my software with the settings you want.

 

[Armysurplus]

Fish on LOL You have My attention. Is there any documentation I can read, before I ask some dumass questions?

 

[neurosurveillance]

Fish on LOL You have My attention. Is there any documentation I can read, before I ask some dumass questions?

No, please ask your dumbass question. I will do my best to answer them.

 

[Armysurplus]

My first question is how many pixels are needed for a good ID.
I'm sure a facial data base would be required, would I need to build one or is there one built I can add too?

 

[neurosurveillance]

My first question is how many pixels are needed for a good ID.
I'm sure a facial data base would be required, would I need to build one or is there one built I can add too?

Before face recognition can be executed the face has to be detected.
The face classifier has been trained several face features, like the shape, eyes, nose and mouth. The less features are visible, the less confident is the classifier. If you are standing right in front of the camera with a clear sight, you can cover parts of your face and the face detector will still catch you, because all other features shine out strong enough to give the classifer enough confidence.
So, the smaller the face becomes, the less clear are its features and the classifiers confidence. It can't give you an exact pixel number or distance threshold.

For face recognition a classifier gets trained specific faces (by taking picture with the face detector) by the software user. To archieve optimal results the trainings images should be taken in the same environment condition as in those the face recogniser is activated.

 

[Armysurplus]

I skimmed a little on the subject today, That is what I thought take as many variables out of the equation for best results. I didn't want to let the cat out the bag this early, but. I have a SD6CE225U-HNI coming from Andy. Will This meet the hard ware requirements? SD6CE225U-HNI

 

[neurosurveillance]

I skimmed a little on the subject today, That is what I thought take as many variables out of the equation for best results. I didn't want to let the cat out the bag this early, but. I have a SD6CE225U-HNI coming from Andy. Will This meet the hard ware requirements? SD6CE225U-HNI

I am using ONIV. The Software works on windows. These are the min hardware requirements. With your camera you can archieve high resolution images which one the one side will lead to very accurate results. On the other side it demands high processing power so If you are using an ten year old laptop you will have a low fps rate. The processing results will be good though.