How much accurate is Real Sense face recognition?

[TGome1]

Hi guys,

I would like to know how much accurate is Real Sense face recognition? I want to implement it for a corporate user registration and identification, but I'm not sure how much margin of error has.

Thank you!

Bests,

Toni.

[idata]

Hello Tonicanog,

Thanks for contacting Intel customer support. The face recognition feature is in a beta stage, this meaning that it does not work flawlessly, Additionaly there is limited support for the face recognition feature. In the following link you will find an example of face recognition feature. https://software.intel.com/sites/landingpage/realsense/camera-sdk/v1.1/documentation/html/doc_face_face_recognition_data.htmlLet me remind you that a new RealSense device will be realsed soon (January 8th, and the pre-order is now) and some of this information may not be applied to the newest RealSense devices. Thanks for your understanding and interest. Best Regards, Juan N.

[TGome1]

Hi Juan,

Thank you for your response. The solution we want to implement doesn't have to be to much secure. I'm talking about a user login / registration to show interesting products to the user in a retail store.

We have to identify users correctly but data stored will be not to much private as a corporate information for example.

Do you think it can work for us? We are thinking about if we have a lot of users registered, like 1 million users, it will be difficult to differentiate them.

Regarding the new Real Sense device, will be a depth camera like the SR300?

Thank you again,

Toni.

[MartyG]

Toni,

Yes, the new D415 and D435 cameras are depth sensing cameras.

If you are planning to use face recognition with the new cameras and their RealSense SDK 2.0, you would have to combine the SDK with a face recognition system provided by a third-party platform such as OpenCV. For more details, google for 'opencv face recognition database'.

Facial recognition databases are also possible with the older SR300 camera. It may not be suited to the scale of project you have in mind though. For example, the in-built face recognition support in the old Windows-based SDKs ('2016 R2' and '2016 R3') can only store 20 faces per database. To store more, you have to create multiple databases, and the more databases that the recognition has to check, the slower the processing becomes. OpenCV may therefore be able to provide a more elegant and efficient mass-storage solution.

I would also recommend using a professional database solution such as MySQL for your RealSense face database storage instead of simplistic text-file based databases.

[TGome1]

Marty,

Great! Thank you so much for your recommendations.

Just one last question. I understand that I will not have problems identifying people with OpenCV platform.

I want to be sure if now a days is posible to have a 100% verification between a lot of users registered.

Thank you,

Toni.

[MartyG]

The confidence level of a verification can be increased, I believe, by increasing the number of facial landmarks that are checked when comparing a face to a stored database image.

I discussed this with another user earlier in the year.