#167 Matt Powell: The Impact of AI on Security and Surveillance

31:50So change is very quick when it comes to this side of the AI house. Yeah. And what are the largest categories of deployments? Is it uh security is it safety uh is it uh in factory settings or hospital settings is can you kind of break down your your customer base so the largest that we see is um it depends on complexity right so So Mexico City, we have Mexico City. That is 60 ,000 cameras. I want to say it is. So when I was talking to earlier about running and, you know, knowing if people are running or fighting or whatever it may be, I think that has 10 ,000 of those modules. so it can kind of track people wherever they go um it's got facial you know i think there's a thousand um license plate recognition a thousand um facial recognition um and then you've got five command centers you've got a false alarm ai when people are calling in that if they're doing false alarms you can train it um so that it recognizes um you know false alarm calls that come in and And then you've got 13 ,000 kiosks around the city that all have two-way video so that you can push a button and they have a 24-hour medical center where people, you know, if they're on the street, they can push the button and it pops up somebody.

33:37There's 30 full-time people 24-7 in this medical center that sit in front of a camera and can have a two-way conversation with you in 13 ,000 locations around Mexico City if you have a medical event. That is a gigantic and that is a hybrid cloud system that we run down there. Then all of a sudden you back off and you've got, you know, some factories that they're even more complex. So they may only have 100 cameras, but they're running 10 to 15 of these modules on each one of them, detecting all kinds of safety things that are happening. So their notifications are, you know, in the complexity of the data that they're putting off that they go through when they look for, you know, safety related is a very different thing.

34:30You know, so we see cities a lot. We do a lot of sports venues. So we have our occupancy counting and, you know, how many people are coming in and think, you know, we have. a bunch of vehicle things. We have our under vehicle system that scans the bottom of a vehicle and then uses AI to understand if there's any type of modifications to the undercarriage of the vehicle and sends an automatic notification that it's trained with grenades and bombs and it has a magnetometer to detect changes in metal alloy. So if you look at like FIFA, we did 40 of those units. But the amount of data that they're putting off is massive.

35:19So it really just depends. If I had to rank it, I would say cities are the largest commercial settings where people want data on what's going on in their environment. Manufacturing facilities is extremely, those become pretty large. hospitals are very big because we're we're transitioning into a world of virtual care and so virtual care um allows us to pop up a camera in every single room and then know if somebody has fallen out of the bed um know if somebody you know if you have patient elopement where somebody is walking out of an area that they're not supposed to um so you know you go into a hospital or a hospital chain and now you've got, you know, 500 rooms in the hospital, you multiply that times, you know, 50 hospitals in a hospital system, you know, you're looking at large amounts of cameras, large amounts of data, which is why being on-prem is so important to be able to process all of that on-prem, trying to do, you know, this isn't for four cameras.

36:25You know, if you're doing, you know, two, three hundred cameras at a time, you're going to end up with a large amount of data that you need to process on site to get immediate notifications of what's going on. So really just depends. You go from 60 ,000 in a city to where you go to, you know, 100 cameras in a manufacturing facility or, you know, a farming facility. But the amount of data that it puts off that they go back through and they audit is incredibly impressive with what it provides you in terms of data. And are these systems getting cheaper? I mean, certainly the hardware is getting cheaper.

37:07Models are proliferating. Is it getting to the point where consumers are going to be able to deploy this kind of video analytics?

37:21it's a it's a good question um that it is definitely getting um cheaper in terms of how many things that you see out there that promise stuff and it's getting less expensive as well um but i would say coming to the consumer um so for businesses it's getting a lot less expensive. To do some of this stuff years ago was cost prohibitive. Recording and video management being completely different. But nowadays, depending, you can go, for example, through a company like us, and you're able to pick out the modules that you want. You can have your integrator build your hardware, or we can build the hardware, You can build the hardware yourself at this point.

38:13And then you can go. We have tons of business owners. They build the hardware themselves, install it. They pay for their service and maintenance over time. So they get all their updates and so forth. And then they're off and running and they're at the same level as when you see on TV and they AI for your business and you think it's for the big guys. It's not. It is completely democratized at this point. It's very consumer friendly. You can get access to it for business. Residential is a little bit different. A lot of companies will not sell to residential. Companies like us, we won't do residential unless you're talking about the governor's house being a residence or the president's house being a residence.

38:57And the reason is, is because the technology is typically going to be above your average consumer to be able to scale it. You know, the amount of customer service challenges that you would get out of, you know, selling into houses all over the place on high end analytics. Right now for residential, you can get basic, you know, motion analytics and, you know, basic things like that. But if you're talking about higher end types of analytics, I don't know if it will ever get to that point where people can do it. There are residential companies that are providing more and more video capability. But the amount of, like I said, scale to release high end analytics into the residential consumer world, the scale is just not there for higher end stuff.

39:57But for lower end, you know, basic security, basic things in your house, it's there today that people can find. Yeah, I mean, I see people talking about, you know, facial recognition locks and things like that. And I've spoken to people about elder care and this idea of outfitting homes with multiple sensors, including cameras, that then learn the patterns of behavior of the person who's living there. and then can alert caregivers or family if there's a change in behavior, whether it's a fall or just a change in behavior that might indicate increasing dementia or something.

41:00Are you saying that those applications still don't reach the level of sophistication of what you guys are doing and that's a cost issue or is it coming? So, you know, facial recognition for door locks is, you know, very different. So let me give you this example. If you put on sunglasses, you know, so, you know, mobile phones today that use facial recognition to access them. If you put on a hat and sunglasses, they start having problems. to get to the point where when people wear sunglasses, you are still detecting 40 points on the face so that if they put on a mask or they put on sunglasses, they can still access an area is a different level of facial capability because it's moved beyond face to pattern recognition and patterns on the face.

42:03And they just happen to be patterns on the face. They It could be patterns in a fingerprint or patterns in your ear because ears are just as unique as fingerprints. So the basics of having your phone turn in your face left and right in order to access is very different than having facial to the point where you can use it for multi-factor authentication. And it's a biometric to get into a highly secured area. So you're going to have that area of the market start to consumerize where people will get that because they will be willing to deal with a lower level of accuracy than the commercial world is.

42:47And so there are companies that will move towards that. And then you'll have larger companies like Amazon and so forth that will start to do that. there have been kind of when you look at elder care, starting with fall management, where people have started to get a lot of startups that got a lot of money pumped into them for that. They sell it into the hospitals. The hospitals start to purchase it. The hardware starts to decrease and the cost of the analytics starts to decrease. And then it starts to be something that you can begin to put into the consumer's home. So it's the same thing when you look at the analog camera and the IP camera.

43:31And when the casinos started purchasing tens of thousands of IP cameras, it made all the components and everything else to the cost point where it didn't make sense to make analog cameras anymore. So everybody started making IP cameras and it started to be larger adoption. So you kind of have to have this ramp up time, obviously, where people are going to adopt certain things. The question will be, is there going to be that market where someone has all the mechanics to be able to go and charge thousands of individual homes and monitor for these things? And all of those, the technology exists right now to put it in a home and tell if somebody's fallen down.

44:18You can train certain behaviors. I mean, we have stuff for farms where we know if cows are going into labor. so you can train it on certain behaviors because cows will start to have a certain pigs and cows have a certain behavior when they're starting to go into labor and so if you own a farm you kind of want to know that if it's happening in the middle of the night and get a notification so you can go out you can take care of it that stuff already exists so elder care type of technology already exists the question is going to be at what point does the industry grow to where companies can support the ability to deploy it.

45:01My personal opinion is that I think you will probably see companies that are out there, the gigantic organizations that put cameras everywhere that they possibly can to gather data. And they will start to find that as a means for gathering the data that is out there because it's very expensive to deploy this stuff into houses. You see it in hospitals, and I know the cost on a hospital to be able to do it. I can't imagine how you would be able to do it in the next couple of years in a residential home without having to add on a bunch of different stuff that really invades privacy so that you can gather massive amounts of metadata and resell it.

45:49That being said, like I talked about earlier, everybody thought that the edge cameras would take five years to get to the point where they would have 60 to 75%. Now people think it'll be about three. So I think this stuff is moving so rapidly that we could see that just speed up. It's eventually going to happen in individual homes to be able to detect this if people want that. The question is the scale and cost for the companies that own that type of intellectual property. Um, it'll, it'll get here, whether it happens in the next three years, five years or 10 years is going to be the question.

46:26Yeah. Um, do you guys, uh, use your system at all for online monitoring? You know, Facebook has invested a lot in, uh, scanning live video feeds or, uh, signs of violence and, and, uh, or pornography or something. Uh, that is not, we do have a managed services group that will process video for people. Um, so we do have that cape. It is not something that we, um, generally do. It is something that, you know, we have a managed services group that has the capability to do things like that, where you could scan live video because that's all video is right. You got a camera out there and, you know, it just happens to go back through your server versus...

47:17But online monitoring is something... It's not something that we majorly do at this point. Our managed services group is... They can take video and they can start to do those things and run the analytics on them and then provide data off of it. But it's nothing that we've really looked at in terms of the scale of people like Facebook, you know, they have the infrastructure to do those types of things is different than the infrastructure, you know, for, you know, a video intelligence company like us that goes out and, you know, we might do cities, or we might do manufacturing plants. When you're looking at that, it's just a different level of infrastructure.

48:03A lot of the technology is the same. So like you said, monitor for violence happening, you know, in online videos or videos that are live streaming. But the infrastructure is very different to be able to accomplish that. When a city comes to you guys, who generally is it? Is it the police department or the mayor's office or who is buying and implementing this stuff? So there's two areas that come to us. A lot of who comes to us is the traffic engineering and city planning. So we have a large capability to do pedestrian analytics. So, you know, and then you're able to extrapolate data from there.

48:57So let's say that, for example, you have a lot of reports of near misses that are happening in a mid block area. Yeah. And so you'll go out there or you have, you know, unfortunate fatality. So they want to know when people are crossing instead of at the crosswalks. They want to know when they're crossing and then they want to get a count of that. When's it happening? And then they want to look for why is it happening? Is it, you know, is it related to the bus schedule? And there's an apartment complex and, you know, the transit bench happens to be right across from the apartment complex. and instead of walking to the crosswalk, they walk right across the street to get to it.

49:36So they'll want to put up cameras or use live feeds and install a module that's going to look for pedestrians doing stuff that is anomaly and look for them anytime that they're outside of it. We have a system called Soffit that will actually illuminate people. It's an intelligent lighting solution. So as they cross the crosswalk, we use AI to drive the lighting so it illuminates the individual as they go across the crosswalk. So a lot of times, you know, the license plate recognition side, cities call, but there's a lot of companies out there that, you know, do license plate recognition. And, you know, they use their automatic license plate readers, and then they're connected into different databases.

50:29So we get calls, you know, for that from cities. We've got a lot of interest right now in that under vehicle system for a lot of the correctional facilities and a lot of areas where they're walking around with mirrors to look under vehicles. But most of the calls that we get when it comes to a city are about kind of building that smart infrastructure to where we're able to connect into traffic lights, run traffic lights. We do connected vehicles. So we connect straight into a vehicle and tell them when a pedestrian is going to be in a crosswalk that they're driving up to, to counting pedestrian actions, looking for traffic backups.

51:11All of these different things kind of go into that intelligent transportation systems area of the cities and city planning. So we deal with a lot of traffic engineering departments that are trying to understand that and kind of build that backbone for themselves, that you can get more of that connected vehicles and vehicle to everything, the V2X, CVX systems. That's the future. When you really look at it, being able to have all these things connect in the cities is going to, we see a little bit of it now, but five years from now, the things that we see coming is absolutely incredible of what you're going to be able to do with AI to rapidly speed up all elements of city planning and then understand reducing pedestrian crashes and fatalities and injuries.

52:12That type of stuff is amazing. That's something that we do globally. It's kind of in our DNA because it all goes back to that skeletal model, right? Going back to 2004 with those first LPR analytics, we've been working with vehicles for 19 years now in terms of neural network training and training our algorithms. Now you take that vehicle and we know what the vehicle is. We We can do the make, the model, the classification. We know a vehicle and its environment. We have the skeletal model. So now we've got that human that's out there. So we start to combine these things where you have this environment that is the roadway, that is the sidewalk.

52:52And you can start to make these systems start to interact with each other to where they're all working together at once. And that's where all those modules start to come together and stack on top of each other. So now you're getting all of that data, but you're also actively you're able to work in the environment to change traffic lights, to illuminate pedestrians as they walk across a crosswalk, to be sending data so that you can understand the dynamics of what's going on and whether you need to change the way that you're going to do your road construction down the road. So, you know, it's kind of funny how it all comes back to those very basic things with neural network training and, you know, the convolutional neural network training that we do.

53:37And now moving into vision transformers and so forth to even get more out of these scenes that we're seeing so that we can do more in terms of the data that we can generate. Yeah. How fine-grained analytics or analysis can you perform? And I'm thinking in cities, you know, there's been a lot of talk about weapon detection and facial recognition. these cameras that you're deploying in cities are they high enough resolution and are the algorithms powerful enough to do some of those things to do facial to get it started so there's four steps to facial quote recognition to start the first step you need 60 pixels between the eyebrows so as long as we can get 60 pixels between the eyebrows, we can go ahead and move towards feature extraction, liveness checking, and so forth.

54:45Most two megapixel cameras will give you that, but you've got to be close to it. So you can set it out to 10 feet. Most two megapixel cameras out to about 10 feet will be pretty reliable in terms of knowing that there's faces out there and go ahead and start the loop to go back and look for the pattern that is in the database that it's looking for to see if it's matching. So once it gets those 60 pixels, in terms of other things, so small stuff and so forth, you know, I've seen some incredible things, you know, speaking outside of our company where you're seeing AI use for cancer detection in, you know, slides, you know, instead of, shipping a blood sample or a blood slide across the country.

55:38Now you can just put it online and then you can run AI and see what you can see. That is not stuff that we do. We're more about the environment in general. If you look at all the different things that we can detect, you've got in there, you've got like carrots and cucumbers. And I asked our R &D team, I said, why? And they said, well, down in South America, we have a lot of horse farms. We want to know if people are giving, you know, are they giving a carrot to a horse or are they giving, you know, something else to the horse? So we've had to train it on what these things are. Weapons detection is very interesting.

56:14We have a study that's out where we had a transit authority that asked us to come in and do all these different detections. and one of them was, you know, using an object to strike an ATM. And the number of times that when you first get started on this, looking at the reason that they came to us was, is that the low accuracy analytics didn't know the difference between a baton and a phone because it was a black device in a hand. And so that is where you have to get into the behavior side. It is not enough if you look at, you know, it's not enough just to do the option or the object. You have to have the behavior behind it.

57:01Brandishing when it comes to weapons detection. You know, that's one of the things that they found over time is, is the person is walking with the firearm. You know, so you're detecting the firearm, but then you have to have really high accuracy analytics for that. But then you also have the behavior around it to set the context. Reason being, you don't want a false alarm factory so people don't trust it anymore. They go, well, every time somebody comes out and they're holding a book that thinks it's a pistol. So two megapixel cameras will get you almost all the time to where you need to be in terms of the task.

57:37And, you know, the worst thing that can happen is, is that, A, you have to put up more of them. Or B, you just have to, you know, bring the activity that you're looking for closer to the camera. the more minute the thing that you're looking for is. So if you're looking for something smaller, you're looking for hands to make movements and do things and so forth, you just have to bring that closer to the camera. But about 2 megapixel will give you what you need, whether it's facial or behavior or weapons detection. But then at that point, what are you doing with it? And that's where the data sets, neural network training time, and the accuracy really comes into play.

58:17So you don't generate false alarms. You get good data that you can make decisions off of.

58:25You said cities are one of your biggest market segments. Is that primarily overseas or is there a lot of that in the United States? primarily overseas um we have a lot of cities that we work with and departments of transportation so forth that we work with but that is not that is its related um technology that is vehicle counts pedestrian behavior pedestrian safety connected vehicle um you know really applied analytics is sometimes what i like to call it you know it's augmented intelligence because we're providing really good data. And then, you know, it's kind of applied analytics. How are we applying these different models that we have through the modules in order to give you data?

59:16You know, we do have cities that we partner with to provide, you know, license plate recognition, missing persons, you know, when it comes to facial in certain areas that may be, you know, areas that, And, you know, unfortunately, you have a lot of people that are victims of trafficking and so forth that, you know, you are able to connect to a missing persons database and then be able to look. But that is a minute portion compared to intelligent transportation and looking at the roadways and what you can do with roadways. yeah wow it's fascinating and and i would guess uh it's a growing business uh and and it's sort of an endless market is that right yeah if you if you don't do one thing i think these days if you do one thing you either got to be you got to be real specific about that one thing that you do and hope there's a big market to grow your business um our ceo al likes to say that whenever somebody says, hey, do we do this?

1:00:27He's like, we've been around for so long, it's probably sitting in the back of a desk drawer. Just pull it out, dust it off, make sure it works. We tried it at some point. So we do a lot, but we look at it as we transitioned. We're a video intelligence company and a data company providing data from what's in the environment. So it's a great business for us, you know, whether it's safety or whether it's operational efficiency, you know, it's growing and the capabilities are growing. And it is exciting to see, like we talked about earlier, the processing power change become more accessible and people being able to, you know, really deploy this to their businesses and their environments and their cities, their hospitals and so forth to be able to make a make a difference.

1:01:20Are there systems that are high enough resolution that can recognize a face at some great distance? Or is that still science fiction?

1:01:43It is...

1:01:48This is one of those weird... It is completely possible. Completely possible to do. It's all about resolution. So, you know, if you can zoom in and be able to get to... You know, so we do something interesting and people are always kind of shocked. So we're one of the only companies that can render a 2D image into a 3D biometric. And so you can take a photo off social media or LinkedIn, whatever it is, and upload it into our FaceX system. And then we can use that as multi-factor authentication because we're not looking at the face. We're looking at the patterns, you know, the distance between, you know, the edge of your eyes and, you know, your nostril and the side of your mouth and the curvature of your cheeks and your chin and so forth like that.

1:02:43those tend not to drastically change over the years by the time it would change you would have an updated photo of it so as gravity affects us all and so forth you know you're going to end up with another photo at some point you upload it into into the system so you know being able to zoom out and be able to view someone's face is a matter of resolution um you know when you get to the basic science of it. That if you can get the resolution from zooming, and that's your biggest problem, if you're using a large telescopic lens to be able to zoom in, if you can get that 60 pixels between the eyebrows, you're going to be able to start the process to look.

1:03:30It's the same thing, but you can take a small little image from social media, you can feed it into the engine, and it's going to be able to do the feature extraction and be able to understand, okay, there's a pattern to this face. Let me work with it. So it's really a matter of resolution. A lot of times these just come down to cost. You know, people get scared of facial technologies and, oh, you're recording my face. And, you know, the amount of storage you have to have to record that many faces, it's just not there. It's just that, you know, when you get to the level of what we do, it's a database verification.

1:04:07I hate the word facial because it doesn't really it's just looking for patterns that exist on a face. But I can train it on the same patterns on your ear. I can train it on the same patterns for a fingerprint. It's just a pattern recognition. So, you know, a lot of it comes down to cost. If you can afford that much of a zoom telescopic lens to reach out and view somebody at distance, If you can get those 60 pixels, in general, you are going to be able to, you know, start to do that feature extraction and look for the patterns on the face. AI might be the most important new computer technology ever.

1:04:45It's storming every industry and literally billions of dollars are being invested. So buckle up. The problem is that AI needs a lot of speed and processing power. So how do you compete without costs spiraling out of control? It's time to upgrade to the next generation of the cloud, Oracle Cloud Infrastructure, or OCI. OCI is a single platform for your infrastructure, database, application development, and AI needs. OCI has four to eight times the bandwidth of other clouds, offers one consistent price instead of variable regional pricing, and of course, nobody does data better than Oracle. So now you can train your AI models at twice the speed and less than half the cost of other clouds.

1:05:34If you want to do more and spend less, like Uber, 8x8, and Databricks Mosaic, take a free test drive of OCI at oracle.com slash IonAI. I hope you find the conversation as enlightening as I did.