Tracking Gaps & Jumps: Technical Troubleshooting | Marvelmind

0:01 Let's discuss in more detail all sorts of irregularities you may see during tracking. On the example of this case, there are 16 stationary beacons and 38 workers. So in this case, only worker number 16 was walking around. Let's see all kinds of jumps or omissions of tracking and the source of it. So let me continue or see. So there was a mission, so it produced around four hertz update rate. But as you see, for so this is around one meter, not one around, but between the dots there was one meter. So over three meters there was no tracking. What's the reason? There could be several. It's even impossible to say without deep analysis what was the reason.

0:59 Now, first of all, it could be and most likely it was an obstruction of stationary beacons serving this area or mobile beacon from the stationary beacons. Since service zones were switched off, we cannot know for sure which beacon beacons were serving in this area. But it could be either this one or it could be even this one or this one. Now the whole point is that during this area, during this time, their mobile beacon didn't see the stationary beacons and the system basically filtered out their measurements or even didn't have them if the obstruction was so strong. The second

1:58 possibility was noise. It's a very noisy environment, and this is why the system may detect this. And when the trilateration is not able to determine the location, basically filters out this raw data. We have a so-called real-time player which can be enabled. And so this is real-time player enabled. And in this case, this will be populated with the data. It will be interpolation data, but very precise interpolation. So it means that if you analyze it not in real time but let's say with the delay of one to five seconds or even once per day, then the system or the data you will be getting will be significantly better. Significantly. But this is raw data. Let's

2:57 continue further. Okay, so it's moving. Tracking is fine. As you see, there's concrete. So most likely it's serving. It's been served by this stationary beacons. Now it's moving and it's been tracked, okay? Let's pay attention to this. So for example, in this case, we recommended that the mobile beacon would be placed on the shoulder close to the edge. Why? Because there is a head and the neck, and the neck would be blocking a large sector towards, let's say, potential stationary beacons in that direction. So this mobile beacon would be able to see only everything on the left and pretty large area. Not, you know, 180 degrees, but pretty large sector would be blocked.

3:56 If you move it closer to the end of the shoulder, their angle of blocked sector would be lower. So it means that their potential tracking would be better. Of course, on the left. Okay, it was left shoulder. So on the right shoulder, there is another sensor attached to this one. So this is why they are both receiving in order to combat their obstruction produced by their neck and their head of the walker itself. So okay, okay, this is the left view on the same. Okay, now you see it disappeared. Why disappeared again? There could be two

4:55 reasons, or let's say multiple reasons, but two major ones: it's obstruction or, with the noise. For us now it's even difficult to detect what was the reason. For obstruction, we already recommended: position the stationary beacons as correctly and as closely following our recommendation as possible and place them about beacons as closely to our recommended way as possible. This is good, but kind of average good. You see, it's still very close to the neck. It must be further in order to make the sector blocking by their neck itself smaller and the quality of tracking would be improved. Okay, the person is moving. The tracking is good. Let's see the next. Okay, so this is how the stationary beacons

5:53 are placed. They are placed well. You see, it's even closer. So there, okay. You see, there was some something. Now, first of all, there was a small jump. The system tried to filter out, but the system didn't have enough data because according to all the data available, the person could move so quickly to this area because there are some limits. You can limit the speed of the person. In this case, you will have fewer of these jumps, but there will never be zero of these jumps. And this is important: element so-called false negative and false positive. If we do too many filters, there will be too many blocks like this. If we relax filters, they will be too many jumps like this. So there's always an optimal

6:52 in settings. In filtering settings and many other settings in the Modem under map settings, which will produce the optimal tracking between too many omissions and too many jumps. But it's not avoidable completely. You can only minimize, and with right settings, you have very minimum number of them. But it's not avoidable completely ever. Okay, let's continue tracking. So the tracking is back and you see from time to time their mobile beacon may be even stuck. What's the reason? Well, remember that we have been getting this data over large distance. So it was like 2,000 kilometers away and we were Zoom connected and viewer connected. So for us, it's very difficult to say: was it because there was no tracking or was it because we simply had

7:51 a very poor connectivity? And we may have very poor connectivity because again due to basic internet delays and some, you know, lost packets of data it could be a gain system. But with high probability, in this case, it was also internet delay. It could be also Windows delay because it was running on Windows machine and sometimes Windows does something very weird. So look at how the stationary beacons were placed. So there was electricity charging. Since the plant works in negative temperatures, customers asked us to remove the batteries, so there's no battery inside whatsoever. It's typical Super-Beacon, but without battery, and it's super big and outdoor. So it's protected against dust and moisture. But since it does not have the battery, so it must always be USB powered. So that's how it was

8:50 implemented there. So the person is moving. Okay, you see there was a jump. But in this case, jump I would guess it's with high probability it was a handover jump, most probably. The tracking was split between this service zone and this service zone, and during the handover, the system dropped quite a few location updates. And you see there was a slightly misalignment. So there was a jump and shift between this and this. So this was our work in progress, so we didn't tune their submaps so that to minimize these kinds of jumps. It's not a big deal for this particular case, but the things could be better. So their submaps could be aligned

9:49 better. Simply takes significantly more time, and it has not been done at this stage just yet. You see, once again, this system disappeared or the mobile beacon disappeared for some time and then appeared back. It filtered out some, but it still had one jump. If the post-processing is applied, this and this would be basically filtered out and a very nice smooth track would be produced, which may at the same time end, filters out these jumps and make their tracking even more precise because it would minimize potential noise that is always there. Because we know that the object itself is not noisy, so it's not jumping

10:47 or jittering when moving. So it's a person, and by knowing this, it's possible to apply post-processing and make it not only nice and smoother but even more precise than the raw data. So this is the environment. Pretty, you see, this is pretty complex area because these stationary beacons are obviously not seen because this will be, but these will completely obstruct this beacon. So let's see whether you will be tracking at all. I wouldn't be surprised if there's no tracking completely, no tracking, because otherwise you would need to install additional beacons here like here additionally and track only in this small tiny corridor. And usually there's a trade-off. You need to decide whether you really want to track with high accuracy in this

11:45 particular small area or, you know, that people may not move away from this. So they may be either this or there and this area. They have no other choice either to be there or be in this small corridor or away. So let's see. Oh, see, you see there was a yellow yellow sign for a fraction the second. System is trying to say: okay, trilateration does work. There are some obstructions, something, so it's trying to detect. Okay, still getting tracking. Maybe the visibility was still there because we asked people to put them pretty high. So it means that when the station becomes high, so it means that they, for example, if the station begins is here and the mobile is here, so they see them from the top. So a potential obstruction, shadows are smaller. But if you put the stationary beacons on, let's say, three meters, then the potential shadow would be larger.

12:44 So pay attention to this and place the stationary beacons accordingly. Okay, some minor omissions, some minus block jumps, but again, very typical behavior. Everything is good. Okay, okay, there was a jump, clear jump due to whatever reason, difficult to say. Most likely noise, some sudden noise. Since we have a delay of half a second or second over the internet, it's even difficult to assign a noise from audio to the jump. But most likely there was a noise which first caused the jump, then it led to some omissions because the system is trying to prevent from providing the wrong data

13:47 when it's able to detect that the data is wrong. Okay, you see now it's pretty poor tracking again. The same sources of incorrect tracking: noise, obstructions. These are the two primary, primary. All the rest could be, you know, pretty minor in majority of cases. Remember that mobile beacon was not placed perfectly, so the guys are just learning how to wear them. So it meant that they could be better if the mobile beacons would be placed, or in this case, one mobile beacon would be placed more properly. But it was placed like it was placed. Now let's continue. So overall, the distance is around 60 meters, and we know that in some areas tracking is pretty poor, simply because

14:48 there's no way to provide tracking because there's so strong obstruction. But it's not this area. It's pretty open area. It could be noisy, very noisy, but in terms of obstruction it's kind of okay. It's pretty open. You see, there are far more demanding areas. Okay, tracking is ongoing. Okay, difficult to say what was the reason for this particular omission. But let's see. You see, there was a slight jump. This is another example of, let's say, of our unfinished job. This is misalignment. So it means that this map and this submap, they're not aligned. So this is called M1 M2. So oranges, as we call them in quotations. So you may align, spend a bit more time and align them better. So this was work in progress.

15:48 So we didn't align well here and we didn't align well here. So the tracking was nice, but when the tracking was moving from this submap, so I really assume and believe that this submap and this submap, so there was a handover zone of two, three, four meters. So at this point, most likely their mobile beacon move from this service zone to the service zone over this and jumped during this time. So basic misalignment. And now it will be moving to far more, let's say, difficult area because there's no way you can track using any technologies when the beacons are not visible. Visible means physically visible in terms of radio, in terms of ultrasound, in terms of light. It all depends on what

16:47 technologies you are using. Since we are using radio plus ultrasound, yes, we do have radio connectivity. It's not direct line-of-sight radio connectivity, but we don't use that linear connectivity for location measurement. We're using ultrasound. So radio we are getting because it's scattered over the building. But for example, if ultraviolet light would be used, literally light wouldn't be able to go through this concrete, you know, because it's just too thick to produce too much loss. And ultrasound obviously is not able to go through these materials. So let's see what the tracking will be when the person will be behind this. And this is why we are always saying: place the stationary beacons far, I mean above, in this case. Okay, there's still okay. There was some jumps. Jumps, the same source,

17:45 most likely there was some noise. You see, there was some, you know, clinking, but the rest. If you place them high, then the chances of obstructions will be lower, not zero. The person may crawl, for example, and if it goes to the shadows of these concrete blocks, there will be no way to track the person. But now the person is okay. Some drop, another drop. And you see this could produce. Okay, now it is producing. It's very difficult to say which one exactly is being blocked. Maybe this one because it's kind of behind. And you see, there are some walls or some obstacles which are higher than the person, and very likely those are producing some obstructions. But since this is kind of in the middle, it would be also possible that some of those omissions are part of

18:44 handovers, handovers between different submaps. So some of them could be improved like this one or maybe even partially this one. But many are not because it's not line of sight, and this is very, very difficult again. So this is taller than the person. So if the person is behind this or let's say between now, the stationary beacons are behind. So these are not yet disturbing. But when it's moving now it's disturbing. You see, so now it's getting the data and getting the measurement. But most likely not a direct line of sight. So this could be once again two things. First of all, omissions due to non-line of sight, due to, you know, blocks placed there. And second, ah, because

19:44 the stationary beacons and the submaps are not completely aligned and there was a jump due to misalignment. So very typically it's not a single thing. It's a combination: no line of sight, noise, misalignment. Okay, you see, there was an omission, a couple of jumps, and some omissions. Now you've got the point, I believe already. When system is not able to measure, we can discuss and agree with some customers. We agree: okay, we show everything. Some customers we show only what we believe is the best to show. But

20:43 we also provide the raw data like distances. So it means that in many cases in post-processing, you may even recover significantly more that is measured here because the raw data is available. And it's possible by having only, let's say, one of the distances to estimate the location pretty well knowing the history, knowing the raw data, their distances, and knowing the IMU data, which is also available. So post-processing, in general, produce significantly better results, particularly because you know more and you know the future for this mobile beacon. The mobile beacon doesn't know the future, but for example, at this point, you already know the future. So this is the future. So it's very easy to produce in post-processing a line which would be very precise, following not the line but the curve,

21:42 following this and the same time not having this at the time when it's measured. It's impossible because the person may move this direction, this direction, or any kind of direction. But when we know already whatever five, ten seconds after, so it's pretty easy to provide very detailed coverage, very detailed tracking. Okay, it's finishing now. The tracking. This is why we placed we asked to place the beacons on around four meters and their mobile beacon was on the neck

22:41 so it was whatever 1.6 meters around in order to minimize the shadows. So I don't see where the stationary beacons would be like this 58 to. They are somewhere there, but for example, if the person goes just next to these blocks and if those blocks were somewhere there in the service zone of those beacons, then the person will be physically behind the wall of these blocks. There will be no tracking. Even worse, in some cases there would be faulty tracking or false tracking. How to avoid this? No, again: place beacon so that if you expect non-line of sight, then you place the stationary beacons so that those line of sights

23:40 will not be obstructed. So in this case, you could do overlapping submaps. Okay, this was too long, but you could build submaps from this area and to this area. So it meant that if the tracking is here, the system is measuring from these beacons and from these beacons. And the system is able in the majority of cases to detect if something is wrong, then it drops out those measurements and takes on release measurements. If measurements are matching, then the system is taking both. If there is obstruction behind, then it drops out measurements from this stationary beacons and using only this stationary beacons. So it totally depends on the case. So this is why there is no solution for all cases. For people, taking it's one solution. For drones, is another solution. For forklifts is slightly different solution. And for particular implementation is always a

24:38 particular implementation depending on what you want to achieve in terms of accuracy, in terms of robustness, in terms of investment, in terms of what you'll be tracking: people or forklifts? Where you can place the beacons? Where you cannot place the beacons? So there are many elements, but all those elements are known. We've done this many times, so we can advise, and you can do it by yourself simply by following our recommendations. So let's finish the tracking, and hopefully based on the example of this particular case, it becomes clear what to do and how to interpret like in this case, how to interpret either omissions or jumps or misalignments. Thank you very much.