Subscriber Discussion

Facial Recognition Product Options?

EA
Emad Allam
Dec 22, 2014
IPVMU Certified

We are looking for a Face Detection software application that can connect to IP cameras.

It is required to build a database including known persons photos, then identifying strangers who enter the building.

Can we get a recommendation for suppliers for such apllications.

JH
John Honovich
Dec 22, 2014
IPVM

If you want to identify strangers based on a datbase of photos, what you are looking for is facial recognition or facial surveillance. It is not face detection.

Face detection is simply identifying and automatically capturing a picture of a face. It does not match it or alert on it. For example, a few IP camera suppliers include built in facial detection, but not necessarily face recognition.

As for facial recogniton / surveillance, make sure the supplier has experience in facial surveillance specifically. The difference is whether one is matching against controlled subjects (i.e., a person stares at the camera), vs a person just walking by (e.g., your request for strangers walking into the building).

I could not recommend any supplier that does this well.

But let me ask: how accurate does this need to be? how tolerant are you to mistakes? How much are you willing to spend per camera / door?

IK
Itamar Kerbel
Dec 22, 2014

AFAIK the 2 most popular solutions are HERTA and Cognitec.

HERTA is a simple solution vs Congnitec which is more heavy weight and from what I understand Herta licensed its algorythems from Cognitec.

Good luck

EA
Emad Allam
Dec 24, 2014
IPVMU Certified

Thanks John and Itamar for your response.

For John, Yes you are right I mean facial recognetion.

To my knowlegge is that the customer is using Avigilon IP cameras, but I do not have accurate answers for the rest of your questions.

All I have is just a request for such a solution.

UI
Undisclosed Integrator #1
Dec 26, 2014

I suggest a company called facefirst out of california. They also license from cognitec but do the processing in the cloud to make a little more lightweight on the server.

JH
John Honovich
Dec 26, 2014
IPVM

Have you used FaceFirst? I've seen their execs make some big claims but I have yet to hear the details of a real production deployment and how well it actually worked.

UI
Undisclosed Integrator #1
Dec 26, 2014

John we currently use and deploy facefirst. We now install it with a national retailer that has it deployed in hundreds of locations. I’ll be frank with you I have used numerous platforms that say they have the analytic and sell it as a feature set add. But none of them do it right because they don’t focus on the core tech. There either a video platform with a bunch of integrations to meaningless analytics or they just add the cognitec engine and think that is all they have to do. Im actually a fan of facefirst because they really did it right. They are a facial analytic company and that’s all they do.

Face first has a fluid and seamless process for adding faces into their DB. Off course there are parameters that need to be followed but with any system there is configuration that needs to be done for it to run optimal. I will not go into the detail as this can go for a bit. But what I can tell you is that faces get captured at the store. They are uploaded to the local server. That image is then sent to the cloud where it is indexed with all the information the user has added ie offense age gender etc. If and when that VIP or Perp comes within line of site of the camera at another location that is running the system it queries the system to see if there and matches. If so it will email alert the associated people on the email alert. As far as the quality of the reads goes...you need to have good enrollment pictures and the cameras that are capturing the images needs to be installed per Facefirst specifications. All in all I would rank them on top of the rest.

JH
John Honovich
Dec 26, 2014
IPVM

Thanks.

Does FaceFirst verify matches in real time, weeding out false alerts? How many of the matches are valid? How many are false / incorrect? How quickly do the matches come back?

Historically, the difficulty has been getting it right quickly, because lots of people look alike (across large public areas).

I'd appreciate any feedback on performance particulars.

UI
Undisclosed Integrator #1
Dec 26, 2014

Facefirst does verify matches in real-time. The architecture is not ideal for some users if they have to rely on their current isp…ie most retailers are concerned with POS data transactions vs Loss prevention. So in this case we typically will supply a 3g/4g router (M2M) in the store. Once the person is captured at the store level that image is immediately processed and sent to the cloud. The only bottle neck id say is the connection strength reliability with the 3g/4g router. I have seen light speed hits in the systems that we have deployed. I can enroll a few people into the system and literally within seconds of us walking in front of the cameras we all get picked up and I get an instant email, text or sms alert with all biographical information of the known individual.

As far as the quality of the matches its all configurable. When you enroll images in to the system it has a scale at which you can measure the quality of the images that are captured. Typically Face first says that it only wants to send alerts when the images that are captured meet 90%+ accuracy of the enrolled picture. That can also be adjusted either higher or lower. Then you can set parameters that if its between an 80-89% hit that it needs to be verified before deletion. The system will dump any captures that don’t meet the data accuracy scale. As far as the implementation of the system its like a science that you have to really get right a few times before you deploy. If you do not get the correct parameters right ie line of site FOV lighting and the person direct face you will not get the reads you’re looking for. One thing that I do have to add is that depending on the width of the entrance/threshold you are analyzing it may take more than one camera.

I found this a very interesting technology because it is in no way like hanging a typical surveillance system. The first few stores we did with face first we had to analyze the way consumers walked into a store. We had to understand when lighting hit the front of the store in a certain part of the day so that it would not create halos around the subject. We had to install cameras deeper into the store to get the correct line of sight. Facefirst also supplies certain tools that are not in a typical integrators bag of tricks such as a laser tool that throws two beams in a certain degree angle to get the perfect positioning at the door.

JH
John Honovich
Dec 26, 2014
IPVM

"I can enroll a few people into the system and literally within seconds of us walking in front of the cameras we all get picked up and I get an instant email, text or sms alert with all biographical information of the known individual."

So the store managers than are getting alerts and visually verifying if the alert is correct?

Any system like this is going to have some level of false alerts. I am trying to understand how significiant false alerts are and how / who deals with them. Any experience with that?

UI
Undisclosed Integrator #1
Dec 26, 2014

When the system is deployed only store loss prevention are trained on enrolling subjects into the system. They do take numerous shots of the subject to make sure the highest quality in images. While there may be a very small level of false alerts they are mitigated by the accuracy threshold level that is set in the configuration. If they do not resemble 90%+ of the image that was enrolled then they are just disregarded. In other words they will not send out any alerts. In my experience the false alerts have been very low. To the point that the end user sees the tremendous value in the case capture and close. In regards to who sees the alerts and manages the alerts , it’s done by the regional directors they administer the system. At the store level the LP can also get alerts if the regional permits. I would recommend to do a review on the system if you guys have opportunity in your test schedule in 2015.

JH
John Honovich
Dec 26, 2014
IPVM

Great, we'll take a look.

Btw, how many perps / criminals have they successfully apprehended using these alerts?

Avatar
CAPTAIN RAVINDER REDDY
Dec 29, 2014

The ultimate challenge for the Facial recognition system is how low is the false alarm rate so that innocents are not troubled by the law & order force and how many criminals were identified accurately to prevent the crime.

I would like to know whether any facial recognition software is available in the market which surveillance professionals can say that we can go for it without having to dwelve too much into the failures of the software.

Capt Ravinder Reddy

New discussion

Ask questions and get answers to your physical security questions from IPVM team members and fellow subscribers.

Newest discussions