Hockey Player & Puck Tracking System | Marvelmind

0:01 Let's discuss sports. We are getting, and we have had a few cases in sports, taking particular in hockey. But let's discuss on the example of hockey: how to track a puck. And we need to touch the subject—you know, sometimes even a philosophical subject—what does it mean to track? For example, I see the picture, but I have no idea where the puck is. So in order to track it, you need to be able to measure the location in one way or another—some physical methods. Let's not discuss some hypothetical neutrino-based methods that would be propagating through the materials like earth, et cetera, et cetera. Let's discuss something reasonably possible. So we are using radio for precise clock synchronization and ultrasound for positioning. So even

1:00 before starting discussing the positioning or the puck—which I don't see, where it is—must be able to receive or transmit signal from the puck, ultrasound signal from the puck. Okay, to the stationary beacons, wherever they're placed. Again, I don't know where it is, but imagine that it is there and it was already caught. So the puck is somewhere there. And we want to track it. Or, for example, imagine that we tracked it before, but now when it's inside, the ultrasound must go out somewhere. Let's make it thicker somewhere and being sensed by the stationary beacons, for example, just for

1:58 example, placed one there, one there, one there. But the ultrasound won't be able to go out from—whatever—from the glove of the goalkeeper. Um, let's imagine that instead of ultrasound, we are using ultra-wideband. Ultra-wideband would be able to come through radio-transparent materials, but not without this distortion in the wave. For example, if it's inside the fist already, then the fist would, and the radio wave coming out, wouldn't be propagating in the air or a vacuum. So their delay will be different. So in fact, the measured location will be not here, for example, but somewhere there or there or there. In some other cases, like optical

2:57 tracking, it's even worse than ultrasound because if it's optically not visible, then there's no way you can determine the location. You can guess the location based on, let's say, previous locations. But as soon as it's grabbed and locked, there was no way you can determine it. But now let's imagine a previous situation when the puck was, for example, somewhere there. Okay, you place the stationary beacons there, on the ceiling. And then the beacon is somewhere inside the puck. Possible? Yes, possible. Good, yeah, it would work pretty well. So you would need to install the microphones, the sensing microphone on the top, on the left, of course on the top, and on the bottom. Because when it's on the bottom, again the same story: you will not be able to hear the sound on the top. You can now. It must, of course,

3:56 well, be protected against ice, against water, against dust. Possible, doable. But let's imagine now that the puck would be somewhere—I don't know, somewhere there—and the steel mobile beacon inside the puck. And then the situation is worse and the station beacon's out there. So it means that the ultrasound must go through all this. Even if, for example, you are not limited on budget, even if you installed many beacons, around those beacons will be still obstructed by—you know, by such a dense number of players. You see, because they will nearly certainly obstruct it. So only when the puck will be flying somewhere openly, then it's easy.

4:56 Then, when they in between, you can use some other methods like sensor fusion and IMU-based. So when it's not obstructed or when it's obstructed, you're using the IMU data. You don't need any reference beacons. IMU data is quickly accumulating the error, but you can settle the error later on when the beacon—let's say, there—the puck is back to this area. So the error will be accumulated. So the accumulated error will be cancelled, and you will be able—so it's a pretty complex sensor fusion, but possible. But it always assumes something. What does it assume? It assumes that the puck will be not closed from tracking for too long a time. For example, more than one second or more than two seconds. Otherwise, their accumulated IMU will be too high, and instead of here, the puck will be, you know, measured somewhere there or there.

5:56 Or whatever, so far from the distance. And when the puck is flying very, very quickly, then it's another story. So you may have location update with, say, typical 8 hertz or 10 hertz. But with the flying speed of, whatever, 20 meters per second. So it means that if I have here—so between the location updates, okay, let me remove these things. Between the location updates, you will have one location here, and then 20 meters per second. So 20 divided by 8, whatever, 2.5 or so. So the next location will be 2.5 meters. So pretty large. So it means that you need a very fast update rate. For example, 100 hertz, which is very difficult to achieve. Possible, mostly by IMU fusion. So in this case, it will be whatever, 20 centimeters between the locations, which is already much, much better. But then it's a trade-off because in this

6:58 case, you will not be able to filter. If there is a sensor fusion, so it may start accumulating the error. So in this case, it will be measuring ultrasound-based position, but IMU-based, it may already move to this. And then it will produce. So overall point and overall message is the following: yes, it's possible. Simply remember what exactly you want to track and what trade-offs you are ready to live with. Let's remove this and let's make the task slightly simpler. Let's make that I don't want to track my puck. I want to track people. Yes, this is much, much easier task. You can employ, install the mobile beacons somewhere in the helmets or somewhere even under the clothing. And then you would be able to track pretty easily by installing stationary beacons on the ceiling and the mobile beacons

7:58 inside there—inside the clothes—yes, it's possible. In this case, abstraction is theoretically possible because, you know, pile-ups and all this competition type of things. But for hockey, pretty easy. So tracking players: yes. Tracking of puck: yes, but with many, many restrictions, many trade-offs, and many potential limitations. Thank you very much.