Artifical Intellegence For Sentry Duty.

[wally]

AI is booming right now and has become very accessible. For the technically inclined take a look here:
https://ai.intel.com/practical-appl...sing-the-intel-movidius-neural-compute-stick/

This attempted home invasion local news story is what really motivated me:
http://abc13.com/its-scary-attempted-burglary-caught-on-surveillance-video/2836001/

All the pieces fell together around Memorial Day and a system I've built as an add-on to our existing FLIR Lorex video security system has been working wonderfully.

I ran the clip I extracted of the bad guys from the news footage through the AI and have uploaded the results to youTube:


I think a system like this makes a very diligent sentry when an audio alert is played on the initial detection -- when the first green box is overlaid. So far we've had zero false detections with our system. Its also nice to get an audio alert when the mailman comes or a package delivery guy drops something off.

 

[CapnMac]

As with all things Security, layers are important.

In this case, calling it "AI" (Artificial Intelligence) is a bit of a stretch. I'd support calling it IA, "Improved Automation," since much of the "connected" home stuff has been lumped under "home automation" as a term-of-art.

Those web-enabled doorbells have come a long way fast. You can get into one for under US$100 (my zmodo was $49 with $8 s/h). The text alerts are interesting, and the streaming video illuminating.

However, there are weaknesses with this.
One is that it's connected to my home wireless system. Which is vulnerable to power loss (I have battery backup--UPS--on my xmiter, but if the service provider's local num is powered down . . . ) Also, there's a vulnerability if the service provider limits or throttles access.

The larger weakness is my phone.
If my phone loses charge, no text message.
If I turn the ringer off, not text notification.
If I sleep through the text alarm, then it's the same as not getting the message.

It's a single layer, with quite a few potential failure points.

Ok, I like it, for when one of the neighbors comes to the door, as I can interact with them (if I get the message, see above; and can talk--like not in a meeting at work). But, it's fascinating that I can hear the fog barking at whoever it is at the door. And the dog is there, not whereever I am at that moment.

 

[wally]

Layers is indeed the key. The guy in the newscast only came away unharmed because he had beefed up his front door so the perps ran off when they failed to quickly kick in the door, probably figuring they'd woken the homeowner and lost the element of surprise. If you watched the newscast interview, he was clueless until the next day when he saw the marks on the door and then replayed the security camera footage which the TV news broadcast.

AI is the standard term like it or not -- like Assault Rifle.

The key to this AI is the near zero false alert rate. In fact one of my motivations to add it to our system was we had friends over for dinner and her RING doorbell interrupted us constantly with bogus alerts -- she had to turn off her phone to finish dinner.

Depends on your paranoia level I guess, but I have a cheap cell phone and a pre-paid plan (averages ~$10/month for this usage) that is dedicated to providing wireless data from my house. It sends me alerts if the Internet or power goes down in addition to SMS messages and Emailed photos should the AI detect persons marshaling on my property. When we are home it is in "audio" mode and uses speech synthesis to tell me where the detected person is. Been running almost two months and it has never false alerted. My AI system is monitoring 9 cameras with overlapping fields of view so there is 360 degree coverage of our property.

I think these systems are a game changer in terms of first line of defense -- a 24/7 sentry that doesn't fall asleep or get bored.

 

[lemaymiami]

None of the principles that guide preventing un- authorized entry have changed one bit... and each layer of security is only a delay against a determined adversary... Now that I've spouted off what I was taught (all those years ago....) our world is changing constantly and that goes for every part of it. Each new weapon brings a new defense - each advance in security is being matched with new methods to defeat it... and the field of artificial intelligence is no exception... even though some of the "experts" might not be old enough to legally buy a beer...

From where I stand, A.I. is just another tool in the drawer - but over time will become more and more important. No matter what a group of I.T. geniuses say (recently a group in the industry publicly declared they'd never work for the dreaded Defense Department - not ever if it involved A.I..... -such foolishness....). In short ---- if it can be done, it will be done. The day when security systems are not only directed by A.I. but also will include weapons directed automatically isn't science fiction any more (just speculation -that last bit but probably not much of an exaggeration). Not long after that we'll have our folks on the ground dealing with those kind of threats....

Glad I'm old enough at age 70 that I won't be facing that world -but you can bet that not that far off our young troopers will be....

 

[JeffG]

The far reaching effects of an EMP discharge (accidental or otherwise) become deeper with each electronic device relied on. Make sure your physical barriers for security are in place alongside the electronic.

 

[lemaymiami]

Amen...

 

[wally]

If you are worrying about an EMP attack you need a tinfoil hat! If/when we suffer an EMP attack its TEOTWAWKI

I'm concerned about mundane threats like Home Invasions, Burglaries, Driveway Ambushes and other things an early warning system would let me actually do something to affect the outcome!

 

[Jeff White]

There are a lot of other things bedsides EMP that electrical devices are vulnerable too. There is no one system that will make you safe. You need defense in depth and you need redundant systems so that you are covered if one element fails.

I was recently in a facility that had a state of the art multimillion dollar security system with multiple layers, everything linked together. Network problems ran the security force ragged for several hours responding to false alarms, access badges reading properly but not popping the door locks, alarms not showing on monitors at the control desk but showing on the handheld devices the roving patrols carried, but the roving patrols were unable to clear the alarms from their devices.

Morale of the story is the more complicated you make something, the more there is to go wrong and when things do go wrong they can affect other systems making the operation very difficult.

 

[JeffG]

With EMP devices the size of hand grenades already in existence, and the looney open border style the Dems want for us..... China is smuggling skids of fentanyl into the country, how hard would it be to smuggle an EMP into the US? Find a different way to insult site members, please...
https://www.theregister.co.uk/2009/02/12/electropulse_grenades/

 

[SharpDog]

One of my "AI"'s:

Works even when the cell phone is dead. Because of the local mesh "Dog Bark Network" I get warnings when visitors are beyond visual range.The dogs are inside the house to prevent tampering.

Note: "Dog Bark Network" Copyright 2018, patent pending

BTW please support your local animal shelter, adopt.

PS.I do have Nest cams with IR sensors and zoned movement detection linked to Alexa and other home automation which functions much like the OP described, However, I don't depend on it.I DO depend on my dogs.

 

[.455_Hunter]

I approve of the UA 571-C Automated Sentry Gun...

https://www.google.com/url?sa=t&sou...EQo7QBCCQwAA&usg=AOvVaw0cMm9hqLLXO_m7HCS1YbZP

 

[wally]

I submit that you have selective memory about the false alarm rate of your "Dog Bark Network".

Do you actually get Email or SMS alerts from your "cams with IR sensors and zoned movement detection"? I realized the general need for systems like what I'd put together when we were having diner with some friends and were interrupted so often by false alerts from her RING doorbell that she had to turn off her phone to finish diner.

Y'all aren't understanding that the AI greatly simplifies the system instead of complicating it. No motion zone setup, no light level rate of change adjustments. I've put the finishing touches on a simple single camera system using only a Raspberry Pi ~$30 "hobbyist computer" and a Pi Camera Module ~$25 or USB camera. It can only do one image about every two seconds (the AI takes a massive amount of arithmetic, so its really slow without the AI co-processor ~$80) but it is still surprising useful.

Network problems ran the security force ragged for several hours responding to false alarms

The whole point of the AI is to greatly reduce the false alarm rate by only responding when a person is in the image. I was amazed at how well one of the simplest publicly available models actually worked. The "million dollar security system people" have missed the boat, as this AI has been available to professionals since 2013 or a bit earlier, it really hit the mainstream last year with "easy to use" tools and models from Google and Intel. The "Maker" community is on top of it now, I'm not the only one to have put together systems like this.

If any geeks here are interested, PM me and I'll send you a link to github when I "share" the sample project.