"We have artificial intelligence. We don't have artificial emotion." - AI, Safety + Parking

In this episode of Parking Live, hosts Jade Neville and Matt Darst explore how AI and smart video technology are redefining what a car park can be.

From streamlining traffic flow to identifying safety risks in real-time, today's parking infrastructure isn't just concrete and signage—it's becoming a dynamic, data-driven environment. Our guest, Jonathan Marshall from March Networks, breaks down what AI really means in a parking context, shares practical examples from across the sector, and offers clear-eyed insights about balancing innovation with privacy, trust, and responsible tech adoption.

Whether you're a parking operator, policy leader, or curious about how machine learning is making curbside environments safer and smarter, this episode is packed with grounded, real-world insights you won’t want to miss.

👉 Got thoughts or questions about the show? Hit us up at: https://www.linkedin.com/company/parking-live/

Let’s keep the conversation going.

 Topics Covered:

• What AI really means in a parking context (no hype)  
• How video analytics are enhancing operator awareness and efficiency 
• The balance between innovation and privacy 
• Real-world examples of smart parking in action 
• Future-proofing infrastructure with data-driven design 

🎧 Credits:

Parking Live is a production of Modaxo, passionate about moving the world’s people.

• Hosts & Producers — Jade Neville and Matt Darst  
• Executive Producer — Julie Gates  
• Producer — Chris O’Keeffe  
• Associate Producers — Cyndi Raskin 
• Recording + Mixing — Patrick Emile  
• Brand Design — Tina Olagundoye  

⚠️ Disclaimer:

The views and opinions expressed in this program are those of the guests and do not necessarily reflect the views or positions of Modaxo Inc., its affiliates or subsidiaries, or any entities they represent. This production belongs to Modaxo and may contain information subject to trademark, copyright, or other intellectual-property rights and restrictions. This production provides general information and should not be relied on as legal advice or opinion. Modaxo specifically disclaims all warranties, express or implied, and will not be liable for any losses, claims, or damages arising from the use of this presentation, from any material contained in it, or from any action or decision taken in response to it.

Chapters

• 0:00 Introduction and Episode Overview
• 0:17 The Evolution of Smart Car Parks
• 1:02 Guest Introduction: Jon from March Networks
• 1:41 Understanding AI in Parking
• 3:02 Safety and Security Enhancements
• 5:13 Thermal Cameras and Fire Detection
• 6:25 Reducing False Alarms with AI
• 8:21 Prioritizing and Managing Alerts
• 11:36 Privacy and Responsible AI Use
• 14:34 Midroll here if needed
• 18:53 Combining Data for Better Management
• 25:45 Getting Started with AI in Car Parks
• 27:52 Common Mistakes and Myths about AI
• 30:23 Final Thoughts and Takeaways

Transcript

0:09

Hello and welcome back to Parking Live. I'm Jade Neville, and as always, I'm joined by my co-host, Matt. Thanks, Jade. And hi everyone. Great to have you with us. Now, today's episode is gonna take us into a world that's becoming more and more central to parking mobility, the nexus of parking structures, technology, safety, and artificial intelligence. That's right. Car parks aren't just concrete spaces for vehicles anymore. They're evolving into smart environments. From managing traffic flow to keeping people safe, technology is changing the way operators think about their site. Yeah, and it's not just about efficiency, it's also about trust and safety. Whether it's spotting risks earlier, supporting staff with the right information or making sure customers feel secure, the role of smart systems is growing fast. And it's a complex topic as well. So today, to help us unpack all this, we're joined by someone who really knows the space. Our guest is Jon from March Networks. It's a company that is at the forefront of video solutions and AI driven insights from parking and beyond. And Jon's gonna help us demystify some of the buzzwords and share real world examples, as well as look at how operators can get the most out of these tools while still respecting things like privacy and responsible use. Exactly, and whether you are a parking professional, a technology enthusiast, or just curious about how AI is shaping everyday places like car parks, this is definitely an episode for you. So, Jon, welcome to Parking Live. It's great to have you here. It's great to be here. Thanks for that welcome. So let's start simple. For people who aren't tech experts, what does AI in parking actually mean today? And how is it different than the old days of just having cameras kind of monitoring a parking lot? Well, I think brings an additional dimension to those cameras and it brings some, almost an additional pair of hands and another set of eyes really to the people that are looking at those cameras and working with that technology and working in that environment. Basically, it allows them to be in multiple places at the same time. We are not quite at the sort of the Minority Report where everything's all predictive AI, but it certainly helps us when we are going back and we are looking for areas of interest. Helps us to find things a lot more quickly, and there are certain cases where it can help us to spot problems and address them before they appear. Just building on that a little bit, Jon, where do you, or what do you see some of the new ways that this technology can help car parks over the next year or two? Are we looking at safety measures? Are we looking at occupancy monitoring? What, how can it impact car parks? It's all of those and some more. So if we take the safety aspect first. Then clearly, you know, having high quality images. So as we move through the landscape of security, every iteration of cameras delivers a much better image quality, better light conditions, and allows you to see a lot more detail than traditional cameras would generally deliver to you. So from a safety perspective, we're able to have, again, that extra set of eyes around the building, around the environment. We're seeing a lot more out there than we did with traditional cameras. Building on top of that, we can also extract metadata now. So within the background, we can start to pull metadata. And when I talk about metadata, I talk about things like, what's the color of your top, what's the color of your bottoms? Do you have a hat or a backpack on, you know, and we can even do gender and age as part of that. So the camera is doing a lot more than just projecting an image. It's collecting data about that environment. And why is that important? Well, that's important for a situation where you are trying to forensically analyze or find something that's happened. You know, somebody's had their bags stole in their car park. What's the first thing they remember? Well, it was a male with a red jacket on. So we can go and search for that much more quickly than just trying to find, scroll backwards and forwards, to extract that. So it gives us another level of information there. From a safety perspective, it allows us to monitor areas, which may be, you know, things like stairwells are traditionally the places people hang out. But I have had use cases where people want to monitor high buildings, so lots of car parks, top floors, whatever, are people just hanging around places that, you know, contemplating the unthinkable. So, you know, with some of the algorithms that detect whether people are loitering around there, we can get an alert a lot quicker than just noticing somebody on a camera. Brilliant. I dread to think what it thinks about my age when it sees me walking around. I don't want it to, don't wanna be predicting my age. I like to know what it thinks. You and me both Jade. They are, surprisingly accurate. Some of the algorithms. Well, you mentioned safety and one concern we hear a lot about is car fires, especially electric vehicles. How is technology being used to kind of help spot those problems sooner to keep people safer? So that's a great one. And you know, with the advent of thermal cameras, we can monitor the bays, we can use the thermal cameras, and we can get an alert when a temperature rises above a certain level. Now the thing with electric cars, as you know, you stick it into a very high power charger, it does generate a lot of heat without actually doing that. But there are also systems out there that can detect smoke as well as that heat signature as well. So it's not just the heat signature, it's bringing together that smoke detection as well to say actually, it's not just a very hot battery, it's actually a real fire that's now occurring. And, in real world terms, right? So everyone, I mean, most people have Ring doorbell, cameras nowadays and get alerts. I get overwhelmed by alerts like, someone's at my front door. It could be a cat, it could be a pigeon, it could be a leaf or a spider, right? So how do your systems help staff know when something's wrong without overwhelming them with too many alerts? Yeah, I mean the way that we do that is that we tend to desensitize the camera to a degree. And there are two elements to that. One is that traditional motion detection just relies on a bunch of pixels changing. And the algorithm is very simple. Did so many pixels change, therefore there is motion and that's exactly what you're seeing with the ring doorbells and the cameras there. You'll see a significant shift in the number of pixels and their activity. With cameras that have more of a metadata concept built in more of an AI element built in. It knows what a person is and it knows what a vehicle is, so it's only when it's seeing a person moving that it's actually gonna trigger that. And it's only when it sees a vehicle moving that it's triggering that. On top of that, we can identify areas so if you look at any outdoor camera that has trees in its view, that'll just generate alerts all the time, 'cause trees, foliage, that sort of thing always moves. So we just mask out some of those areas, desensitize them, and start to reduce those alerts. There are also technologies that I get involved in in sort of defense and critical national infrastructure, which is very specifically trained to look for only people. As an example, I work on a site up in Norway, and it's wild. It's rugged. There's wildlife there, there are very little people there. And we've implemented with a partner an AI solution that can literally detect the difference between a deer and a person crawling. So again, all of these analytics really help us to reduce the false alarms. And what does that mean to the people running the systems? Well, it means that when they get something presented to them, they've got a pretty good chance of knowing that that's a real alert rather than the alert fatigue that we get on a lot of systems where it's just, yeah, that's going on in the background. Yeah. Yeah. Not interested. It does that all the time. Is these systems really help us to focus on what's important. So Jon, it's interesting that there's this ability to de-emphasize certain images. I'm wondering, are you doing any work to prioritize certain images over others? Like how do you take this data and ensure that the most important warnings are the ones that get put to the top of the queue and don't slip through the cracks. And that's where, again, setting them up properly really comes into play. Really investing the time to install and configure and tune these. That's what's important. A lot of organizations just put the cameras up. They get a lot of alerts. As I say, it's ignored. So we spend time with our customers when we're working with them to help them say, well, what is the exam question that this camera is designed to answer? What are you trying to find from this? And we will work with our customers to get that tuning just right so that we don't overwhelm them with false alarms. But we are also very clear that what we're bringing to their attention is what they need to see and what they want to see. So it's a little bit about the technology having the capability to do that. But it's also bringing back the human element in terms of working with the customers to understand that and using our knowledge to configure that correctly. And there's a common theme here about preventative measures, trying to catch things before they get to the point where something actually happens. But if something does happen, how can this kind of technology help teams quickly figure out what's happened and then how to respond faster as well? So once you've got an alert, obviously, then you need to, the technology is one element of this. You need to have some robust operating procedures that support that as well. So it's not just having that camera there and those images there. What you need to do is make sure your organizations are properly trained. And that they're equipped to know that in these situations, what do you do? So if we are looking at that forensic analysis, again, if we go back to the earlier example, okay, show me within my car park exactly where somebody with a red jacket and a black backpack was. So you'll be able to see exactly where they were. You need to make sure again, that you are preserving that information in a way that it can be shared. So there's lots of video images and often, having the ability to see something from multiple angles as well, can help you understand really what happened. Because sometimes you see one camera, one flat image, you can draw one conclusion. But if you see an image from another angle or a third image from another angle, sometimes you can see all the things that that first camera can't see. So in terms of that, it's making sure you've got robust procedures, you've got governance around that, and people know what to do in the event of an incident when it's been flagged up to them. And then you come to, is it forensic search based on that, or is this something real time, which I need to respond to really, really quickly? I think you're speaking music to Matt's ear talking about data analysis. Metadata and he's definitely a data man, so I'll let him take the next question. Absolutely. Yeah. Thanks Jade. Jon, I, one thing we talk about with data all the time, right, is the need to respect privacy. And you mentioned loitering kind of as a potential case study, right? And, how do you determine what's innocent loitering versus what loitering might have more malicious intent and then segregate those to ensure that we are respecting the individual's privacy at the end of the day. And that is a great question and that's probably much more challenging to answer than understanding the technology. So we have an algorithm that detects people just sort of hanging around in a space, you know, is a person in the space for this, but it's more about behavior. And, we developed this algorithm, loitering algorithm very specifically for bank ATM lobbies. So, particularly in our U.S. market and our Canadian market, it gets very cold. So if you have any vagrancy or homelessness, they have a tendency go and find a nice ATM lobby. It may not be warm, but it's actually outside of the wind, and they'll tailgate people into that. And then the behavior becomes very different to a loiter. So it's somebody's there. Now, if you go back to what we talked about with lots of pixels moving around to detect motion, somebody goes to sleep in an ATM lobby, there's no motion there. However, our loitering algorithm, very much tracks says that's a person I know that's a person there, that person hasn't moved. But I still know that it's a person there. And after a preset amount of time, we'll get an alert out that says there is a person there. And that's was developed as I say, for ATM, but you know, it's a great algorithm for stairwells, for example, for all those corners. Every car park's got lots of little corners. That, you know, might look great on a rainy night. So as long as you've got some coverage there, the algorithm is there to detect people in that particular environment. You could also apply it to slip and fall, for example. Has somebody entered an area? Have they had a slip and a fall? Are they unconscious on the floor? Somebody's not moved for a certain period of time, get an alert out. And again, that's where I think these tools really help be that third set of eyes. Second set of eyes, really. If we talk about governance, then governance is incredibly important when you're talking about personal data, the security of that data. So what we've done is we have the technology to solve many of these problems, but I've created a five step governance plan, which helps organizations understand both the implications, the benefits, and the risks of the governance surrounding the use of AI within any business, any responsible business. Responsibility, the responsible use of AI and technology is a big question. You know, it's something that we get asked all the time. So what kind of rules or safeguards, could you just give us a couple that should be in place to make sure that this technology is being used responsibly. So I think the key thing there is you've got to have an understanding of how the AI was trained. So if you've got a true AI, it's only ever as good as the data it's trained on. So, lots of things have been trained on, on basically the internet, which gives a whole polarized view of the world, right, wrong, mad, crazy. All of these things are feeding into the model of what it sees the world is and how it makes decisions. So you've gotta understand really how your AI has been trained. You've also got to decide, you know, is this genuine AI or is this machine learning? And there is a subtle difference between the two. Genuine AI is where it's really making a decision. Machines learnings is still applying a set of rules to some data to create an outcome. So that's one point. The other thing is, I always talk about is that whilst we have artificial intelligence, we don't have artificial emotion. So an AI does not have that emotive level that you or I have when we see some data. I once read an article that said, if a statistic causes you to have an emotional reaction, you should immediately challenge that statistic. AI can't do that. We have to do that. So we have to make sure that we understand that the information it's giving us, an AI is giving us, we need to understand how it arrived at that. We need to have some cross checks, checks and balances to say, does it feel right? Is that right? And we need to constantly review that. It's not a one time thing. Also we have to respect that if it's giving us any personal information, that we have to secure that, and we have to make sure that that is secure within our business, and we have to make sure that that is protected within our business as well. Heading in a little different tack here, does it matter where AI is hosted? Does it matter if it's on the cloud or if it's back office? I'm wondering how that plays into some of the things you've just explained to us. Well again, you know, GDPR is something that concerns all of us and if we are feeding an engine with data, we have an obligation to understand where that data is going. And what's happening to it. What's the life of it? What's the shelf life of it? Is it deleted immediately? Is it stored? And from a geographical perspective, you know, is that data leaving borders? Is it crossing borders? And I know I work with some very large financial clients who are very, very concerned about how AI is both implemented, but also what impact does that have on reputation if it gets it wrong. And what is the perception of their customers if they are aware that AI is being used? And, I think it's important if you are using any form of AI within your business, that you need to be very, very transparent about it. You need to be transparent about why you are using it. You need to be transparent about what the outcomes are from that. And something that's really important is ensuring there's no bias in there. Again, you know, based on sample data sets, AIs do have the ability to create bias within output. So I think those are things that we have to be very careful about when we are looking at implementing either an on-premise or more specifically, an off-premise solution. This kind of technology excites me quite a lot because I mean, the speed of how it's been implemented, the speed of adoption, how quickly it's growing in terms of what it can do and how quickly it learns and how we're applying it and how we are learning how to apply it. It's all very exciting stuff. And then the insights element and how it's helped us break down data and understand records and, you know, the bigger picture, you know, of what some of the information that we sit on every day is actually telling us is really interesting. So how does combining things like video and other information, like entry records and payments, help make car parks safer or easier to manage? So again, another great question. As we bring together all these data sources, it's making it a lot easier for us to get to the seat of the issue. And I know one of the key challenges within the parking industry is things like the use of false plates. So whilst the vehicle is in false plate, registered leads without paying the owner gets that. So by being able to tie together both live video and the LPR data as an example, and the payment data for example, we're able to very quickly go straight to that video. We don't have to sit through scrolling for hours and hours and hours of video to try and find that. We use a technology called Searchlight in our products, which links together payment data, LPR data, or any data source that can be time tagged to bring together in such a way that we can just, for example, show me all the occurrences of license plate AB 1 3 XNZ, and we will bring together very quickly all of the video associated with that. So, again, you know, lots of systems take still as the car comes in. Maybe that doesn't quite give you who's driving. So by linking that with other cameras and real time video, you can follow the car around the car park and see who actually gets out. Somebody may say, well actually no I was not driving that vehicle. Therefore, you know, false plates. And we are very quickly able to get to the video and say, actually no you were, because here's the entry, here's the car entry. Oh, but look, that's you getting outta the car there. Rather than just having to take it at face value that okay, we don't have a face we can't prove. So that helps us in terms of if we are managing the car parks, it helps us really get to the root of fraudulent behavior. But also if you're a customer, then you've got at least the degree of comfort that the technology is also helping you when you are in the right, you know, so it really helps that customer relations aspect. If somebody can say, no, no, actually that wasn't, you know, here's the image of you there. And I thought, well, that's clearly not me. Here's an image of me that's clearly not me. So I think that on two levels, it helps the operators, but it also helps the customers as well. It's that two way street of building trust, isn't it? Between the technology, the end user, service providers, you know, and everyone, anyone in between. Yeah. Yep. Yep. I think there's an element of convenience there as well. And for instance, you know, having nice wayfinding signage or good lighting, those are things that improve the customer experience. Are those areas where you've seen AI focused at all? Yeah, I mean, one of the products we have is, it's an analytic camera that does people counting, queue length, dwell times. So it gives you that, those statistics that sit around, for example, your payment machines. Suddenly you've got a queue, you've got 10 people at the payment machine we can alert out that says actually there's 10 people waiting there. Clearly there's a problem with that payment machine. And we can get somebody down there really quickly with that. You know, we can look at things like heat mapping. So where are people within your vestibule areas, where are they hanging around? We can look at counting, so how many people in, how many people out, you know, which entrances are used more, which floors are used more. We can do all of that using the counting cameras. And we can present that information in some very nice dashboards, that really help you to understand both, develop and drive the development of that building such that it meets customer needs better, understand where there are issues, and as I said, you know, if you've got a lot of people suddenly building up around a payment machine, clearly there's a problem. Get somebody there. And it's always much nicer if somebody just turns up and says, yep, I know there's a problem. Let's sort it out, rather than you're pressing a button to ring somebody somewhere. So again, I think by bringing some of those technologies to bear, it helps you from an operational perspective, but again, it drives a better customer experience. I guess you'd also see maybe other types of actions, kind of sending signals that there's a problem. Maybe someone's in queue like by themself for a protracted period of time at a kiosk or, maybe the gate hasn't been raising and you've got a queue of vehicles there. It's really interesting. Yeah. And one of the things we can do as well with a lot of this technology again, is, you know, basic stuff. If a machine fails, we can get an alert up on and get the camera up on that machine very quickly from an operator desk. The other thing that we do as well is, you know, our products are enterprise wide. So you can have a central control room, which has access, seamless access to all of your facilities, so that you don't have to have somebody in the local building who's a technology expert, but you can direct them very quickly to where they can go and solve the problems from a central control room. And particularly in a 24 hour environment, it's important that you've got a good overview of all of your facilities and your estate. And you've got the technology that tells you where the problems are, not quite before they happen, but as they're happening and allows you, as we said, to deal with it. What about measuring success? How do your operators know if these tools are really making a difference without collecting more data than they actually need? And again, I think that that's a very simple thing in that, you know, you need to benchmark, measure your performance and measure your occupancy before, measure your occupancy after. One of the things that I talk about in my AI governance is be very clear about what you're trying to achieve. Are you trying to achieve cost reduction, customer satisfaction, efficiency, you know, all or all of those things, very, very, very clear about what you're trying to achieve. Set your benchmarks and measure it, and don't be afraid. Stop using it if it doesn't work, or at least go back and revisit some of that governance. You know, we shouldn't be doing anything in this world without, I hesitate to use the word business case. But you know, it needs to pay its way. Some of these technologies can be expensive. We need to be able to demonstrate a return on investment across customers. Whether that's cost reduction. So what are your baseline costs? What's it gonna do? And be very, very clear about that in an organization before you embark on this journey. If a car park operator wanted to get started with this kind of technology, where would you suggest they begin? Great question. As I said, you know, they need to understand what they're trying to achieve. So the first step is what is the exam question we're trying to answer here? You know, what do I want my business to achieve as a result of this. Once we've got that bottomed out- so again, we've gone nowhere near technology yet. Once we've got that bottomed out, we can sit down, we can talk and say, well, how are we gonna achieve that? What are the issues we're going to face? What are the risks of doing that? Reputational risk is a big, big issue for AI at the moment and the implementation of AI. So again, you know. AIs that make bad decisions, that's a reputational risk that reflects very, very quickly and be very aware that the media are very, very quick to pick up on this. There's a number of organizations that trumpeted the fact that, you know, they were able to reduce staff and efficiencies by using AI. And with starts, slowly starting to see some of them actually reemploy and rehiring on the basis that A, it didn't deliver what it was supposed to do, or B, it wasn't creating the right reputational output. Once you've got through all of that, then we sit down with our customers and we discuss what's the appropriate way to do. And one thing we are very passionate about, at March Networks, is that we very much like to do a proof of concept with our customers. So, you know, we are happy to fund technology and some of our time to help customers understand by putting technology in, helping them understand what it does, working with them to sort of vet it in and test it. And then at the end of the period saying, this is what we've achieved. Did we set out to achieve what we set out to achieve? Yes, let's move to a much better rollout, a bigger rollout, and more staged approach. And what are the, well, I suppose, are there any common mistakes you see people make when they first try to use these systems and how can they avoid them? So the common mistake is, assuming that everything will be, you know, roses on day one. It's not going to be. As much as an AI has to be trained, people have to be trained in the best way to leverage those technologies. They have to be trained on how to get the most out of it. And as we said earlier, if you don't have the fundamental business principles of operating procedures, none of this is going to be any use because you're getting information out of a system. You need to understand what are you gonna do with it. What's it telling you, what's it telling you to do, and what's the outcome that you're gonna achieve as a result of that? And that's a really interesting kind of AI myth that AI proponents bandy around, that it just kind of does it all on its own on day one. Are there other AI myths that you'd like to explode while we're talking to you? Like, I think, some of the people who may be less enthused about AI might say, oh, it's prohibitively expensive. Or, it's too out there. It really doesn't understand my business. What are your thoughts on those and other myths? I think that they, there are a lot of myths out there. I think that there are a lot of enormous benefits you can get by doing some very simple things. And as an example, I was talking at the parking conference in Scotland last week. One of the presenters was a lady who was responsible for one of the the councils up there in the parking. And she had adopted a very simple tool, something that we all have access to most of us called Power BI. And she was using that. She'd taught herself with YouTube how to use this and was developing, you know, really quite impressive results by analyzing her data very simply, using a very simple tool. So we all know about Chat GPT and all of that, that's great, but there are some very simple tools that we're starting to see creeping onto our desktops, creeping into use that we can all start to use. They don't take enormous amounts of training. We do have to invest some time and some effort in learning how to use them, but you know, they can start to help really ease your workload. They can make your working day a lot easier by taking a lot of the, some of the the more mundane tasks off you, of data analysis. It's very easy to get started on some of these tools and to get it to do some analysis for you. And the more you use these tools, the easier they become to use. Jon, this has been a really fascinating conversation and the AI conversations are almost the hardest ones to try to wrangle into one back and forth kind of podcast in a Q&A 'cause it's so easy to go off track 'cause it can go in many different ways. But before we wrap up, what's the one thing you hope listeners will remember from today's conversation? Take it seriously. Treat it as a very serious undertaking. Understand why you're doing it. Be very clear about what you want to get out of it, and be ruthless about how you measure that. As a data person, that's music to my ears, Jon. That was awesome. Thank you so much for joining us today. It's been my pleasure. Thanks for having me. Thanks so much, Jon.