Talking Data and AI: Smart Technology for Search, Rescue and Security

TALKING DATA AND AI EP 2: Smart Technology for Search, Rescue and Security

This webinar originally took place in 2025, all information was correct at the time of recording. 

We’re joined by Doug Lothian, CTO at Zelim, who’ll be sharing how they’re applying AI to uncrewed maritime search and rescue – from drones on the water to intelligent surveillance that supports coast guards and navies across the globe.

It’s such a powerful example of how data and AI can be used for good – protecting lives and making a real impact.

Zelim is a leader in AI-powered, uncrewed maritime search and rescue technology. Our product portfolio spans from Uncrewed Rescue Vessels (URVs) to AI-enabled wide-area surveillance and protection systems (WASP), delivering comprehensive solutions for maritime search, rescue, and security operations.

Transcript

SPEAKER_00 0:00

Welcome to the second in our talk and data AI webinar in the series. It's great to have so many of you on the call today. I'm your host, Lyle Ritchie, head of Talent Solutions here at TED Resourcing. For those of you who don't know, Ted Resourcing is a digital IT and change recruitment company. I'm delighted to be joined by Doug Lothian, CTO at Zellum. Where Doug will talk to you about the AI journey Zellum are going on. For those of you who don't know, Zellum are a technology solutions business with a mission to make unmanned search and rescue the industry norm. Before I pass over to Doug, just a couple of quick points. Doug will talk for around 30 minutes, and there will be a 15-minute QA session. So please, if you could thumbs up the questions you want the answers to. And there is a chat box which you can engage with throughout the webinar series. Perfect. So really looking forward to this, and I will hand straight over to Doug. Doug, over to you.

SPEAKER_01 1:05

Brilliant. Thank you, Lyle. And um thanks for inviting um Zellum and myself, obviously, to talk. We've been um friends of of uh of um Head for a long time. Um uh and and I'm sure um Hugh Um uh at Head would would tell you lots of stories going back 20 odd years. So I've not head for a long time. Um really pleased with uh to work with you guys um again at Zellum. Um so what we're gonna talk about today? We're gonna talk about smart smart technology for search, rescue, and security. That's really what what we're about um at Zellum. Um we help search and rescue, defense, crews, energy companies um achieve faster detection, alerting, and tracking of threats to life and safety of navigation. Now, what does that mean? Um effectively, if someone falls in, we're there to help find them when they fall in the water, either during the fall element or when they're in the water. Our systems can do that. And that can be from an aerial asset, um, that can be from a platform or indeed from a keyside. Um, but equally, um, the asset itself um often needs defended. And you can just look and see what's happening today, um, especially in Ukraine, uh, with the Ukrainians attacking the Russian fleet, with the Houtis attacking um uh carriers in the Red Sea, etc. Actually, vessels and platforms are now also becoming targets. Um, that's primarily happened in the defense realm. It's now starting to happen the commercial realm. It may also start to happen um, unfortunately, in the future in the kind of leisure realm, and we'll touch on a little bit about that. Um, our solutions, they they exceed US Coast Guard detection standards, and I'll touch a little bit on that. They exceed international man overboard standards as well. Um, and of course, it's a maritime environment, it's quite a chaotic environment, so they have to work day, night, and in poor visibility conditions. Um, as a company, um, we've partnered with global naval um uh um it's not companies, but uh but global navies, um, as well as Coast Guards, as well as um different energy organizations, both on the renewable side and indeed the oil and gas. Um oil and gas was a dirty word, but um I think uh we we're all um uh now acutely aware that that's um it's something that um we also need to look after the people and the assets out there um just now. Um so um what are the challenges? Um just looking at it, I mean I've mentioned a little bit the the global insecurity. Um in half year two, 2024, there was about 80 incidents um uh because of missile strikes, drone attacks, or hijackings um of commercial vessels. Um, um, predominantly, this was actually in the in the Red Sea with the Houty rebels. Um we've also got huge numbers of people crossing oceans, and they're crossing oceans to escape either uh poverty um uh for economic reasons, or also because of um uh obviously instability in their regions. And they're crossing across the globe. This isn't something we often hear about this in the English Channel, but it's actually happening across the globe where people are crossing waters and unfortunately um getting into trouble. Over 200,000 crossing um the Mediterranean or the English Channel last year. Um we've also got a lot of people offshore. Um, so you don't realize how many people actually work offshore. So about uh nearly 2 million people um working offshore um on assets. Um, some of those assets are fixed assets, like platforms um in the North Sea or or um the Gulf of Mexico, Gulf America, I'm not sure what you want to call it nowadays. Um uh, but also um just looking at they sometimes have incidents. So um I would ask the question um uh how many commercial vessels um have incidents each year, but our marketing people put it in here. About 5% of the global fleets of commercial vessels bang into each other or bang into something each year. So that's about 3,100 vessels have a collision. So every vessel, um, according to statistics will have an uh will have a collision in a 20-year period. 65% of those collisions or collisions are actually attributed to human error. So although they have lots of um systems on board, um uh those systems um uh aren't foolproof. Um and indeed we had an incident um in the North Sea recently, um, obviously, where uh uh um a vessel uh collided with a carrier um just off Grimsby. Um there was also an incident um uh where a platform was actually struck recently by a support vessel going at high speed, um, and unfortunately both of those are down to human error. Um passengers, there are a lot of people going offshore on holidays. Um, some of you may have been on cruises yourself. 2023, there's 31 million passengers, and that's growing. The the cruise company, if you look at Royal Caribbean, if you look at Disney, um if you look at um Norwegian cruise light, all of these companies, MSC as well, all of these companies are building um larger, more fantastic vessels. Some of these vessels are really uh floating palaces with lots of uh um let's lots of entertainment um solutions on board. Nearly 40 million passengers are predicted um by 2027. Um the UK MIM MAIB um uh actually estimate there's about a thousand man overboard incidents each year. That's not of the cruise ships. Let me just underline that. That's not a thousand of cruise ships, they estimate a thousand man overboard incidents each year um uh globally. So that's about three a day people fall overboarded to the water from leisure, from commercial um or from um uh um cruise. Um quick challenge for you guys, and and if you can, please answer some of these uh some of these questions in the um uh in the in the chat box and and Lyle will try to answer uh um or read them out. Um does anyone do anyone know um what's the maximum survival time in the North Sea? Um so if you felt in the North Sea, in the clothes that you're wearing today, so not in a dry suit, not in a wet suit, but actually whatever you're wearing today, what's your maximum survival time in the North Sea? Um if you can answer that, we'll go back to it in a second. Um second question According to the US Coast Guard, if you had a SAR helicopter, a search and rescue helicopter, so one of the big um red and white uh S92s or AW189s flew over you, what's your probability of being seen if it flies directly over you? And then lastly, how good are we, we as humans, um, at spotting and recognizing and tracking things in a chaotic environment? The sea is obviously very chaotic um uh with movement um everywhere different um uh uh from waves breaking to floats them and jets them to simply the the different waves. How good are we at spotting that? We got any answers for that?

SPEAKER_00 7:36

Yeah, we do, yeah. So just the the first question ranges from 30 minutes, 40 minutes. Scott's put two hours. Um so yeah, that's the kind of range.

SPEAKER_01 7:48

Around that range.

SPEAKER_00 7:49

Yeah.

SPEAKER_01 7:50

Interestingly, um uh so for that one, um a lot of our um a lot of our our beliefs um are based on on what we've heard from the Titanic disaster. Um and actually hypothermia can set in relatively relatively quickly, but actually the survival times are quite lengthy. Um, in a 15 to 15 degrees Celsius um water, um which uh it's approximately 14-15 degrees Celsius just now, just in some of the shallow areas of the North Sea, um, you've got about a 17-hour, 17-hour um life expectancy survival time um uh in the North Sea. Now that can vary, different people, different fitnesses, different uh um uh uh body sizes, etc., can change that. But that's the maximum. So it is quite lengthy. So so long as you have a life jacket on, you can survive. However, if you do not have a life jacket on, that's when things start to get difficult. The 30 minutes is a very interesting one. Um around 80% um uh of um deaths occur in the first 30 minutes, and that tends to be people who are not wearing life jackets. After around 27 minutes or so, um, the cooling of your muscles actually stops you from swimming, so you end up with um with swim failure. Um uh, and there's a number of a number of deaths that occur because of um uh cold water shock as well, but your mat from survival time 17 hours. According to the US Coast Guard, yeah, helicopter?

SPEAKER_00 9:07

Yeah, so I I've got here 5%, um yeah, a couple of five percent, yeah, six percent.

SPEAKER_01 9:16

It's it's actually interesting that that people think it's so low. Um and it it is low, it is low. It's it's around 18%, one eight, eighteen percent. So it is low. Um uh but they've got um uh amazing kit on board, but unfortunately there's always a human in the loop. There's usually a human, and unfortunately, we as humans, and that comes down to the next question, we as humans in the loop are not very good at spotting things in a chaotic environment. We're very good at spotting moving things uh in a static environment. So in a wood, we're very good at spotting a deer or spotting a rabbit in a field or something like that. But actually, if things are if lots of things are moving, we're not very good at spotting those things. Right. Um this is this is a real life um incident. This was actually uh there's a film and a book came out about it. It's not without hope. It was a number of um US um college football players went on a fishing trip, and unfortunately, uh they their vessel capsized overturned. Uh they were in the water, I believe it was like three or four days in total. Um, and unfortunately, um a number of them died. Only one person survived. And from his survival story and from the the actual accounts of the Coast Guard, um uh we can see that they're actually on the right hand side you can see um all of the search patterns that were conducted by the Coast Guard. And you can see many of them are over are overlapping. Some of those are from um vessels, some of those are from um aerial assets, so helicopters and fixed wings. Um and interestingly, they were overflown multiple times um uh and sometimes multiple times in a 24-hour period. So actually, we're not that great at spotting things in the water, and that's really where we we came from from a Zoe perspective. So, Zellum, we built, we developed a product called Zoe. Um uh God, let me jump back. We developed a product called Zoe, and uh um we we designed this product to try to support search and rescue, to try to find that small thing in the water, that really small thing, which is a head in the water. Now, finding people in the water um isn't something new. Uh, over 80 years ago, um, the US Coast Guard um started a study to look for um interesting submarines and vessels. It was during World War II. Um, and that study was looking for large things in the water, and that's became the basis for what's called search theory um um in uh um uh in the modern search and rescue world. Now, search theory relies on things called lateral range curves. Lateral range curves are those curves on the on the right-hand side there. It looks like a normal distribution or a bell curve, but actually it's it's it's defining on the x-axis um the distance laterally, so left, left and right, um, that you're able to detect a person in the water and the probability of detecting that person. So you can see laterally here, you've got 0 to 0.1, 0.2, 0.3, 0.4, um, and then probability up the left-hand side. It's a very narrow window that you can theoretically spot somebody 100% of the time, and then it tails off um as the person um goes away, um, or as the as the person uh uh ranges away from the the um asset. As I mentioned, that's theoretical. Um, in reality, down the bottom you can see here the probability of being spotted, even if overflown, is only 18%. So there's a difference between theory and and um uh and real life operational practicality, and that's because um uh we have to consider all of the different variables. So particular target objects have different detectability indexes. They they may be different clothing if it's a person, it might be a different life raft, um, it may be a different size of object um spotted from different distances, and the environment changes, the conditions change. So different sea states, um, visibility can change, all of those things affect it, and that's why operationally um the levels are so low. Um interesting stat there. Only 19%, or actually less than 19% of all searches for a person in the water actually yield a saved life. So it's it's imperative that um number one, if you do go out in the water and you play in the water, you have a life jacket. Number two, you have some sort of communication or signaling device. But actually, equally for our search and rescue um providers, that they have a solution that allows them to improve their detection capabilities. Um and as I mentioned, that's why we developed Zoe. Zoe was looking for this small thing in the water. We looked at classical detection um solutions and we looked at artificial intelligence. And this was way back in 2021, 2020, 2021, and really before artificial intelligence was um a buzzword. We started gathering data in the water of people in the water, and we were using um different solutions, um, different types of infrared cameras as well as different um uh daylight or RGD cameras. Um we then went on an exercise actually with the UK Coast Guard in 2022, and we started working um a little bit closer. The UK Coast Guard and the US Coast Guard happened to be on that uh um on that exercise. US Coast Guard took interest and we ended up with a contract with the US Coast Guard in 2023. It's a cooperative research and development agreement, um, which are really to look at um the employment of artificial intelligence to detect, uh detect, recognize, and track objects in the water, and also to find a scientific measure um uh um to or scientific way to measure detectability. Um, it sounds like a very simple thing to say. How do you measure detectability? But um until um uh there are various methods, is probably the easiest way to do it. You can get into a lot of theory um uh talking about this, but detectability is a very difficult thing to measure. Um so we've we've developed the solution in conjunction with the US Coast Guard, with our CRADA, but also with UK DASA or DSTL. So that's the defence and science uh um accelerator um and the defense science technology labs down at Botton Down. So we've worked with both these guys to develop the solution. We have it deployed um operational, it's in offshore energy, oil and gas and renewables crews, um, and we are working with um with our defense partners in in various navies as well. Um how good is Zoe? So I mentioned the 18% probability of detection. Zoe actually offered a five times higher, so 96% probability of detection from double the distance. Um and one of the key things here is actually looking at distance. Um so we've five times higher, double the distance than traditional mechanisms, and we've got a very low false positive rate. Um, false positives for all you AI um uh people will know um is uh is uh can be a killer. It's a very balance between assuring your true positives and your false positives. The picture on the right hand side um is actually a true incident. This was the Hendrika vessel that listed in the North Sea. Um it listed in about a sea state eight, so it's about a 14-meter wave. Most of the crew were rescued. Unfortunately, it became unsafe to hoist off a the last crew member, the captain, and the captain was actually asked to enter the water. He entered the water, um, and unfortunately, the the crew were so busy in the aircraft um uh flying the aircraft and avoiding various other things, they missed the person entering the water. They were picked up later, um sometime later, um, but um it wasn't an immediate um uh detection tracking and pickup. We ran Zoe on the same footage that were supplied, and we detected that person in the C-state 8, 40-meter waves within three seconds of them entering the water, and we were able to track them. So had Zoe been on board that helicopter, um, uh that person wouldn't have had such a long wait, and potentially um it could have ended much worse. So lateral range curves, um uh very narrow um uh band for higher detection capability. And the dots down the uh down down the bottom picture there shows um the white dots um uh show the successful probability of detection or successful detections, the black dots show missed detections. So in a search swath, an area that that's the the asset is searching, they are missing targets. And this is actually taken from um the uh US Coast Guard's own literature. Zoe, on the other hand, um, through the full extent, the full field of view um of the cameras that Zoe is deployed on, we have a 96% probability of detection, which means that we don't miss those objects. So if we are searching for an object or searching for a person in the water, we've got a much higher probability of detection and of finding them, which should yield more lives saved. But actually, um uh in all of this, um yield you lives saved is is the human aspect. There's always accountants, and there may be some accountants on the call, there's always accountants that are looking at the numbers. Um, this also means you've got a far more efficient search. It means that instead of having to search an area five times to then have a high degree of probability that you you have a high degree of coverage, um, you can search it once and have the same uh degree of coverage. We worked it out on the incident that I mentioned earlier, the Not Without Hope incident, and they could have saved $1.5 million in uh search asset use if they had um had Zoe on board and working working um in that particular search. So it's quite a saving on a single search basis. It also means that if if if an asset doesn't have to search an area five times, if the person's not in that area, they can be off searching in another area, which again means that you're using those scarce resources more effectively and efficiently. So that's one aspect of Zoe. So Zoe were originally designed to help find people in the water, a very small thing in a chaotic environment. And really that's where the MOB site started, started off, looking for a person in the water. We were then asked by a number of customers to say, well, could you actually find a person if they fell off a ship? Could you find it earlier before they hit the water? And that's where our MOB solution comes in. We're able to detect a person um as they fall. So that one second that they've fallen from, say, a deck before they hit the water, we detect that. And we detect that to the international standard, the ISO 21195 standard. Watchkeeper. Um, this is where if we could find little things in the water, well, actually, we can find bigger things as well. So uncrewed or autonomous surface vessels are currently um uh causing um havoc, obviously, with the with the Russian fleet. There's no reason why that couldn't be turned against a friendly fleet as well, or against um leisure or commercial assets. Um Watchkeeper and Shield, these two solutions, um, offer the capability to find small low radar cross-section objects before they become a threat to you or become you become a threat to them. So Watchkeeper looks forward and makes sure that we don't bang into things or or um uh and that can be static object or dynamic, and shield covers the whole vessel to make sure that um perhaps some malicious objects don't come and attack you. So look at them independently. Um board detection, I've touched a little bit about this, but we use the artificial intelligence to um detect both the falling person um uh and and obviously um alert to that falling person. Now, many solutions um simply um uh do frame by frame detection, detection, detection. We can't do that in this environment. We have to look at the behavior of the falling person as well to ensure, first of all, it is a person, and secondly, they're behaving like a falling person would look. We then uh will alert, but we also have the capability based upon pan-tilt zoom cameras to slew that camera onto the location of the person in the water and to track that person using the uh using the pan-tilt zoom of the cameras. Um so again, using the artificial intelligence capability and the tracking capabilities of Zoe, we're able to not only detect the fall in the water, track where the person is in the water, but because we can also then determine the latitude and longitude in the water, we can simply send a rescue boat straight to them. And that saves um uh um a lot of time and effort, obviously, for the for the rescuers, but also more frequently will yield um uh a life saved. Um that system is designed again to 21195 man overboard detection standard, and that requires that we have about 95% um detection capability in our own tests. Um, in an operational environment, we far exceed that. Um uh and we are we are um uh close to 100% um uh detectability um throughout the range. It also um integrates with uh with uh the systems on board. There's no point in having a standalone system. This system has to integrate um neatly with uh with systems aboard your asset, be it vessel or platform. Moving on, watchkeeper. Watchkeeper 225, as I mentioned, looks forward um for any any um uh potential hazard. Um the term tends to be live wide area uh moving imagery. Um and really what we're looking for is collision avoidance. We're looking for things that we may bang into or things may bang into us. We mentioned that uh humans are great at spotting things that are moving, especially in a static environment, um, but they're rubbish at spotting things that are static in a moving environment. So actually, some of the things that get banged into most often. Are things like navigation marks or anchored ships as happened in the North Sea of Brimsby. So we're very good at spotting things that move, not so good at spotting things that don't move. So we spot uh local radar cross-section vessels. Now that could be leisure vessels like ribs, or in the case of obviously the migrants crossing, the the rubber dinghies that they're crossing in. Also spots dynamic or static hazards. That could be larger vessels or smaller vessels. And the static hazards really are your navigation marks that quite often get damaged on entry and exit of ports. We augment that with AIS data. So we can pick up AIS targets and obviously augment that on screen. But we don't need that. We can actually have non-AIS assets and also track both where they've been and where they're going to give a predicted vector for those. That all integrates with the SLUTEQ of the Pantil Zoom cameras and can also integrate with the onboard radar or Marpa systems. So quite a comprehensive solution. That goes forward really into your Shield 360. So looking around the whole perimeter of the vessel, and it's the same idea. It's making sure that vessel is protected and it gives you the threat vector designation, giving you your vector predictions as well as your closest point of approach and time to closest points of approach. We can also designate friend or foe as part of that. And again, use the um uh the electro optical or infrared cameras to salute a cue onto it. Um now one of the questions we often get asked is uh is why do we choose artificial intelligence as opposed to choosing a classical detection solution? Um, classical object detection, I think a lot of you guys will know this, um, tends to tends to require a solution like background subtraction. So effectively you're looking for pixels that change on screen and you're highlighting those pixels on screen. And that works great in a static environment like a desert environment. So in defense, there's been lots of work put into moving target indicators for on land, where you tend to have trees or buildings or rocks even that are static, and you can then look at which pixels change and which pixel change then tends to be a target or a moving target. That doesn't work so well in a marine environment because the sea itself is constantly moving. You have white caps um uh um and you have reflections off the water. The other thing about classical object detection is it may tell you that something is there, but it doesn't tell you what that something is. Our solution allows you to recognize what that uh what that thing is, and that helps us to then designate whether that thing truly is the thing that we're looking to detect or if it's something else. Also helps you detect whether that thing may be a friend or a foe. So we may have something, for example, falling off a vessel, but is it truly a person? If we alert that's a false positive, that can then annoy the crew. If it annoys the crew, the system's likely to get switched off. A diving bird can look very much like a falling person, both on infrared as well as on your electrooptical cameras. But actually, if we then look at the characteristics of a diving bird and we look at the um the uh the the pattern um that uh the the the the the bird makes, we can quickly um eliminate that from our our uh um uh from an alert and and not bother the crew. Equally from a vessel perspective or uh or um uh an object on the water, we can determine if the thing is what we think it is. Um if it's known, it can become a friend. If it's not what we think it is, it can become a foe, and we can continue to monitor it. Um throughout this process, um we we've we've learned a lot of lessons, and I've tried to just basically put them down into a couple of a couple of different um uh classifications here as to what those those lessons are. The first one I would say, and we fell into this back in 2022, we had some fantastic results straight off the bat um uh from our artificial intelligence. Um a lot of this was lab gathered. We had fantastic um uh scores, and our mean average position are F1 scores, and we believed that um we had a solution that was ready for the operational environment. Um, I would say from you from your own perspective, you're doing this, don't be too over uh don't be overly confident with your lab test results. You have to get out into the operational environment, especially if it's a chaotic environment, and test that early, test it often, and gather as much data as you can from that relevant operational environment. Um for us, that has been the theme all the way through. And those two images that you see on the right hand side, I like that. We often uh um are out gathering data in the worst conditions you can imagine. Now that could be throwing, not people, that's actually a dummy getting thrown off a cruise ship. That's a heated dummy getting thrown off a cruise ship at night time um with quite a congested background, um, which just makes everything very complicated for the for the AI detection solution. The image underneath is actually um in the middle of the North Sea, um, and that's a fog bank. Um so that is a fog bank, and the item that's being highlighted is a vessel that's approximately 1.5 to 2 kilometers away from the uh from the platform. So again, nighttime fog bank operational environment, um, and we're we're demonstrating that capability. That's really the things that you have to do. That comes through actually in the items that I have here. So, lessons learned on data collection. You need large, relevant, and diverse data sets. They need to be collected in an operational setting, and it has to be under varied environmental conditions. Um, obviously, we we see cars on the road, we now see um uh cars on the road with with a lot of the um uh automated features. Um, and over over the last five to ten years, uh the cars have been getting better and better, but they're still collecting data, they're still improving because the operational environment is such a difficult environment to work in. Labeling. For us, labeling is key. Um we have tried lots of approaches with automated labeling solutions, etc. But actually, um human labeling is by far our uh our um uh um our most successful approach. We need accurate labeling, accurate, accurate annotations, and it has to be captured with supporting metadata. All of that supports our training. Um we just as an example, we we collect um and annotate in the region of about 120 to 150 frames um uh per month. Um it's a big overhead, um, but it uh it definitely yields benefits. Independent validation sets. Um there's no point in having a great uh a great training set if you can't validate um what's out there. So actually, you need a truly independent, relevant um uh and uh validation test set. We started again uh having fantastic results, but then we found that we'd actually cross um uh cross-polluted, cross-pollinated between our test set and our our uh our training set. Um so we spent a long time um collecting truly independent um data that could be validated um uh um uh in order to assure the results that we have. And that was required both from a regulatory perspective, but also a customer perspective. Our customers um have been very rigorous, as you can imagine, in assuring that that um uh the AI solutions that we have are are um uh are correctly tested and correctly validated. Um and the last thing is your operational acceptance testing. Um, there is no point in just testing during the summer. When you are doing your operational acceptance testing in a relevant environment, um, you have to do it in the worst conditions. We had uh um tests running um during Stormywind last year, last January. So when the winds were blowing at 130 miles an hour um in the midst of the North Sea, we had our system up and running in those environments to prove its capability to validate uh um what we had seen both in the lab and in our tests. So you need extended operational acceptance testing in all operational environments and all operational conditions, otherwise, um what you have is is uh is is really a uh um uh uh a fair weather system. Um quick picture, this is this is a quick video. Um this is Southampton docks, this is on board a vessel. Uh, this is um an example of our our watchkeeper solution, really just looking at looking out for static dynamic objects, etc. Unfortunately, in this thought in this in this um uh video um there's nothing that caused a collision risk. Um, but this just gives an example of uh um of the system speeded up uh the the red funnel ferries don't move that quickly if you've ever been on one. Um this is the vessel exiting. You see the nav marks being being detected there, they're the static objects we want to avoid. We can see a Q ship here exiting, and the skip problems have to keep an eye on that, so we're highlighting that there is a moving object. Objects that are moored, um uh um we tend not to highlight. So because we're taking in the AIS records, we understand objects that are moored or perhaps um uh at anchor, in which case what we tend to do is not highlight those. We only highlight the objects that the engine's running are and are defined as being underway, or static objects such as um uh uh nav marks. Um you can see here uh very little um uh in fact uh um if any false pod is being alerted here, uh we really are just highlighting the objects that the skipper needs to see. And that's just a quick, quick example um of our watchkeeper solution looking out um uh from the front of a vessel coming at Southampton. Um if I jump on, the team that we have, very diverse team. Uh we've got teams that have worked in defence, we've got teams that worked in technology, we've got teams that have worked um in local government, um, and we have a fantastic team. None of this will be possible without both the team um uh um and the senior team, but then also actually the the engineers that we have. We've got a fantastic team of engineers, really enthusiastic and dedicated team of engineers. And that's really what we recruit for. We recruit for the attitude and the enthusiasm. And it's one of the things that I know Lyle um has always uh enjoyed, I think, or hopefully enjoyed, trying to find the the guys that we have. Um now I think I've just hit the 30-minute mark and bang on just now. So I'll take any questions if any of any questions.

SPEAKER_00 31:56

Fantastic. Thanks so much for that, Doug. Bringing Xellum to life. Uh Zoe, fascinating the products you've you've developed, um, I think, and how harsh the environments you've had to test these in as well. Um in uh this this weather. So you know, really, really interesting stuff. There is questions coming through. I'll just kick off firstly, Doug. Going back to Zoe and you know the start of this process and and going through the kind of AI route. I think people will be interested in finding out why did you engage and think about AI then at the start and and that sort of kind of journey you've gone on.

SPEAKER_01 32:33

Yeah, no problem. Um so I mentioned the differences between AI and classical detection. You know, classical detection, uh I don't know, I'm just the the simplest form is the background subtraction. Many of the team have worked in defense, many of us have seen um the um the false positives that come up actually um when you're looking um at some of these solutions using classical detection um in operational assets. Um, I'm not going to mention the the technology of the companies, but um we took part in an exercise um recently in Canada. Um it's called SAREX, it was off Vancouver Island. Uh, there were aircraft flying using traditional search and rescue um solutions on board. They had a huge number of false positives to the point where the operators were ignoring um and simply using their eyes to try to find um uh objects that may be a person in the water. Compare that to Zoe. We took the route because we wanted to not only detect that something was there, but to positively classify that what that was the thing that we were looking for. And that was really where the AI came in. And then taking it further, we could also then add behavioral aspects to it. So we could add a temporal aspect, understanding what what um the behaviors the the at the the the target had exhibited to then again determine whether uh whether that was something that that was of interest to us that we should highlight. We during that exercise um uh we were looking for um two missing people in the water from a vessel um uh offshore. Um we ran that, we detected all of the objects that were that we're looking for, the two people in the water, we detected the vessel also, um, and we had zero false positives during that full exercise. Um so really that that was why we went down the route, um, and that kind of validates uh our our thinkings. Our thinkings, that's not a word, but thinking.

SPEAKER_00 34:17

Yeah, great stuff. No good answer. Um I'll just crack on. We've got lots of questions here. So um how scalable is the Zoe system for deployment across fleets, ports, offshore installations?

SPEAKER_01 34:31

Yeah, it's a very, very scalable system. So um from our uh from our perspective, we we we can deploy in in multiple multiple ways. Um we can deploy at the edge, so we can have processing um uh um uh carried out actually aboard the assets. So if you imagine if you have uh an offshore oil and gas platform or a renewable floating wind platform, or indeed a cruise ship, um, we can actually be processing aboard and providing real-time uh um uh alerts to the to the people on board. Um we can also um uh process um in public cloud, for example. So we can be processing um uh um uh remotely. So again, using LTE or SATCOM links, uh we can have a camera, for example, in a port, uh looking out, looking for uh looking for objects, and we can we can then process um uh uh in the cloud there. Um and we also have our own private cloud for for for for security that that uh that that we can also process. So as a system, um we we would argue that it's almost a you can never say infinitely scalable, uh a lot of us do, but it's it's it's a very, very scalable solution either in the cloud or or at the edge. Um as a system also uh we provide real-time uh over-the-air updates in the same way as a Tesla car would. So as new functionality becomes available, we can offer that functionality to our users um again through the LTE or or or SATCOM um secure links, and they can toggle on or toggle off um those new features um uh as they like.

SPEAKER_00 35:55

Okay. Good stuff. And I think you've answered one of the other questions just regarding do you employ public cloud for processing or storage, or do you process at the edge?

SPEAKER_01 36:05

Um, yeah, yeah, we can do both. Um you can do that. Yeah, and and really the the beauty there is is that um we we we cannot say that we're camera agnostic. I think a lot a lot of um a lot of times um uh uh people say that we we like to we we have a set of cameras that we like to use, but really most cameras today, most electrotical cameras use use very similar um sensors, most most IR use very similar sensors. So we believe that we could be agnostic that way, um, which means that you can almost take your your cameras that you have fitted today, and so long as they can be purposed for the role that that we want them to use, we can then process um, for example, in the cloud very, very fast um setup internal round.

SPEAKER_00 36:46

Okay. I think that moves nicely on to Hugh's question, just regarding what other use cases can be uh can the tech be used for. So canal paths, lakes, rivers, etc.

SPEAKER_01 36:56

Yeah. Yeah, 100%. It it's really because we've we've focused on the maritime environment, there are a number of solutions out there that look at the terrestrial environment. Um we've focused from uh inception on the maritime environment. Um when we're looking at uh at um uh waters offshore, uh you obviously have a lot of uh a lot of movement there. When you when you look towards harbours, when you look towards canal paths, when you look towards reservoirs, um you tend to have less less movement. It actually makes it slightly easier for us to detect the things. Um so absolutely, any deep water environment, reservoirs, quarries, canal paths, um uh and inland waterways, absolutely fine.

SPEAKER_00 37:36

Okay, sounds good. Thank you for that. Um couple more questions here. So, how do you merge the data from multiple sources, say visual and framed, or or do you process each separately and combine the results?

SPEAKER_01 37:49

Um it depends on the situation. So, um, we what I would say is we process each individually and we combine the results, it's probably the easiest. Uh, if you try to do one one or the other, you're gonna end up with um with some latency. So we we try to think of it. The way we discuss it is almost like uh like a uh a hive um uh where you have worker bees, and the worker bees are obviously gathering the data um and can provide a level of processing, and then we have the uh the the queen who is managing it, and actually our our our nexus layer. We call that we call that the the Zoe Connect, which is is the layer that uh that that combines everything and presents it.

SPEAKER_00 38:29

Okay, good. Makes sense. Um and just so Alan Wallace, sorry, Alan Wallace has asked, are there any other providers doing this, or are you spearheading this approach using AI? Um, is it being taken up by companies, crews, shipping companies? So I suppose, yeah, who are your competitors? I know you said about Zoe and the competitive advantage, but is there anyone else doing this type of thing in the market?

SPEAKER_01 38:54

There is there is there's other people. You know, it's a nice thing. From our perspective, it's quite a nice thing that there are competitors because that validates uh that there is actually a market. Um so for the for the man overboard um solution for the person uh falling overboard um and detection, so there's an international standard, ISO 21195. Really, that there's two main companies in this field. There's us and there's a company called Mars. Um, and Mars have a um uh a solution. Both Mars and ourselves are currently ISO 21195 stage two um uh certified. So both very good products, both in the competitive marketplace, both offering comparable solutions. Um we um are the only of the two at this stage uh that also offers the in-water detection. So not only detecting the fall element, but also then detecting the person in the water and tracking that person in the water. Okay. You take you take that step further, you look at the watchkeeper. Um, there are other solutions in the marketplace um offering comparable. So Orca, uh Orca AI offer a um similar solution for the watchkeeper detecting um uh objects that that could cause um uh um uh could cause a collision risk. Um and Orca AI um really have a footing in the commercial shipping um environment. C.ai um are another another um company um and they predominantly um are in the leisure yachting market, but they are now expanding more in the commercial marketplace. You take it a step further and you look at the the the Shield or the um or the Watchkeeper 360. Um solution there, we believe, is is is actually the only solution that provides the full 360d coverage on a vessel. There are other solutions, so the radar solution aboard um does provide 360-degree coverage, and we're not trying to replace radar, we can augment with the radar. Um radar um is is very good for larger objects, but smaller with uh objects with low radar cross-sections, obviously, um uh can be very difficult to find equally high precipitation, or if they turn the gain up because of sea state, then even um uh um uh some vessels can become obscured um in those situations on radar. And that's the beauty of uh of our solution. The last one from a defense perspective is radar is an active solution. So if you're pinging uh um someone with your radar, they know that you are watching you, they know that you've been picked up. Ours is completely passive. So audioelectrical or electrooptical infrared, if it is watching you and tracking you, it is completely passive. Um you don't know that we're watching.

SPEAKER_00 41:23

Okay. So it seems like you're in a a pretty you know competitive place, but you're at the forefront with your with your product um and and driving that forward. So um that looks good. Uh so just cautious of time, um, somebody's asked. Also, we've been talking about Zoe specifically, but what I suppose quite high level, what other products do you have alongside the kind of AI tech within Zelda?

SPEAKER_01 41:48

Yeah, no worries. So we started off, and I think you introduced us as uh as uh uh our aim was to make unmanned rescue the norm. Back in 2020, uh we designed an unmanned rescue vessel. So actually, we we built a full full-scale prototype um uncrewed uh uh vessel, and that uncrewed vessel we had with authorization to run on the River Forth, and we did for really about two, two and a half years. Uh, we achieved the world's first um incapacitated um uh casualty recovery from an uncrewed vessel. Um uh and we then worked with oil and gas um uh company North Star Shipping to design um uh a full-scale vessel. So from the prototype, from the six and a half meter prototype, a seven and a half meter um fast rescue craft incorporating our novel technology for lifting the person from the water. Um uh that was launched about two years, sorry, a year, year and a half ago, was launched. Um, and we're actually building another version of that at this stage, specific for the cruise market. That's a fully uncrewed vessel, um and as I say, can lift a person from the water. Now, a component of that is is uh an unmanned um uh in-manned rescue conveyor that we call SWIFT, um, and that can be retrofitted to other vessels. So this Swift conveyor um can be retrofitted, for example, to pilot vessels or rescue vessels um uh or crew transfer vessels that may be um operating um and may have a man overboard. It's already been um certified as a primary life-saving device, so it's a life-saving apparatus um and it and can be the primary life-saving device, and it doesn't require, again, any human to touch it. So if you have a mass casualty event where you're lifting um uh, say, four or more casualties from the water, um, uh, then a person doesn't have to physically lift, physically lift. Interestingly, we talked about the US Coast Guard earlier. Aboard a US Coast Guard cutter, um, it takes four people to lift one person from the water if they are able. Um, whereas our solution can lift a person without a human touching it. And an interesting anecdote that we're told unfortunately, we're all getting bigger. Uh, we all uh obviously eat too many cakes, eat too many pies. So we are getting more heavy, and when we're wet, uh we get even heavier. There was an instance in the US um where A person had fallen overboard from his leisure vessel, US Coast Guard Cutter came to get him. And unfortunately, he was quite a large chap and was unable to be brought aboard the cutter. They were forced to tie a rope around the person and then tow him to shore to invoke the rescue. Had our solution been there, we would have been able to lift him or multiple people aboard. And obviously that embarrassing event would not have happened for the person.

SPEAKER_00 44:24

Okay. Wow. So yeah, lots going on there then. And thanks for describing some of the other projects you're working on. I think it would be worth uh noting if if anyone wants to find out more, please visit Zellom's website and on LinkedIn Zellm's page as well. Talked a lot about the events that are on and the products they're developing, more information there. So that's it. Thanks to you all for joining today. Doug, a massive thank you to you for um talking to us um about Zoe and Zelm and the journey you're on. There's lots, lots of information in there that I'm sure um people on the call have taken. So thanks again for that. Um just a quick one from me. Um I will be hosting uh the third um episode in our series um on the 2nd of July with Gary Crawford from Owen Dale Advisory Services. Um he is going to be looking at when intelligent software understands the user on the 2nd of July. So please um join me then. There will be a break in the summer, and we'll kick this off again on the 27th of August. So please look out on our socials for that. Thanks again. See you all soon, guys.

SPEAKER_01 45:26

Thanks, everyone. Thank you.