Help: Analytics That Detect Person Wearing Cap Or Dark Glasses And If A Second Person Comes Into A Restricted Area

Identify the moment when a person jumps into a restricted area where there is already a person working. 

So, Felipe, if there are 2 people in a defined area, you want to trigger an alarm? Is that right? If so, I think there are solutions, think of it like a queue line application where the queue is defined as 2. Main thing I worry about is whether there are any times or scenarios where another person comes into that area that would be legitimate. That would be hard for analytics to handle.

Second, they want to detect and alarm the main control center if someone wear a cap or dark glasses, do you think is there a possibility of detect this?

There are some forensic tools that claim to be able to search for such characteristics but I don't recall any mainstream offering claiming to alert on them.

I am curious what others add.

Hi Felipe,

Both your use cases can be achieved. below is how you can do it.

1) For scenario one you can apply crowd formation. Here you can set threshold of 2 which means as soon a 2nd person is seen there is a trigger.

2) Training a network for hats/sunglasses can be done and below is screenshot where it detects person wearing a red hat. Person with Hat

PM me if interested and we can discuss.

John, This is Suresh from Duranc. I earlier posts i was asked not to promote and was  answering to the post that things can be done and how they can be done. Duranc platform has certain capabilities today that can be done but didn't disclose as we were told not to use for promotional purposes.

If the member is interested and wants to contact us he can always approach through you as you know who we are or the poster is.

Any custom object detection involves more details and requirements and commercials cant be defined from a single post.

Having said that crowd formation is part of our analytics offering and $25/camera/month.

Regards,

Suresh

Hola Felipe

Iss offers a specific analytics for missing helmet (I believe this is used in the mining industry). Not sure if this can be tweaked for your application.

Thank you 

Gianni

Excellent, Thanks a lot Gianni...It may work, 

The question as always with analytics is not can something be done, it is "can something be done reliably." How mission critical is accuracy? I have been involved with analytics for over 12 years, and I would never claim anything close to 100% accuracy on detection of a particular event or trigger. I am a sales engineer with Bosch, and have deployed hundreds, probably thousands of channels of analytics. The biggest problem is managing expectations. The CSI Effect. We have very advanced analytics which perform very well in most circumstances, and are free in the camera. No additional license, no extra charges. And when someone wants to try a novel use or a tricky application, we supply a demo camera to setup and try in the real world, on location. It is irrelevant how it performs in the lab.

If 100% accuracy is needed, analytics is most likely not the answer. I always advise my customers that analytics job is to say, "You said if this happens, to let you know. I think it just happened. You need to look now". Its job is to reduce the amount of video data that must be reviewed, but it still requires what I call the "human Filter".

So finally, whatever technology you are investigating, "Try before you buy!".

You might be able to do this with American Dynamics and CCURE with a facial recognition add-on.

I imagine you could setup an analytic rule that confirms a face is detectable (no hat, no sunglasses), if not, then trigger a CCURE output that could activate an alarm or lock down a facility.

You could also achieve the "jewelry customer escort" rule by setting an analytic configuration that alerts when two objects are more than a certain distance apart. It might also serve to monitor when an escort leaves a customer unattended, which might also be a security risk. 

Thanks Leonard, I will try to ask for a solution to the representatives of the brand down here in Mexico, if any,

Thanks a lot.

@Felipe FD I'm CEO of Camio. If you provide sample videos, we can give you precision  and recall numbers for your particular resolution/camera angle/lighting so that you know whether the solution works that way your customer needs it to. It won't be perfect, so it's a matter of determining whether there's an acceptable rate of false positives and false negatives. For example, customers want us to detect hooded night intruders and Camio search results include false positives from very similar hats and head-related gear: https://camio.com/c/91puhvzvry0w/app (all can be improved, but level-setting in scene PPF like this)

Carter, Thanks a lot for your suggestion, I will ask the customer, because they dont want to send out footage of their installations.

Thanks a lot but I will contact you.

Hello John:

I am proud and happy to belong to this community. I am member since more than 3 years and this is the first time I ask for help, and its good that I got it, a lot of experiences and advices, that's good.  Thanks John and thanks to all community.