I have to learn this for a project I'm working on and it seems... Vaguely ok? Feels like it definitely could be more complex, and hopefully the accuracy is good enough
I have to learn this for a project I'm working on and it seems... Vaguely ok?
Feels like it definitely could be more complex, and hopefully the accuracy is good enough
I think it would have been nice to have at least a few other examples of where KFs can be used. They tried to sell it by saying it can be used for all sorts of things, then just gave the same...
I think it would have been nice to have at least a few other examples of where KFs can be used. They tried to sell it by saying it can be used for all sorts of things, then just gave the same example you always see, robot/vehicle motion. That's like saying "a spoon can be used for so many different and unexpected things, for example, eating soup!" OP, got any suggestions?
Also, these are tough to implement without having good intuition on what these matrices look like in a realistic problem and what each part is doing. When I learned this in class, most of the questions were on this point. I guess the way they often teach linear algebra doesn't really prepare you for intuitively using it in real life problems.
Imagine you have a side-on camera view of a game of ping-pong, and you want to estimate the 3D coordinates of the ball at any given point in time using just this side view. To achieve this, you...
Imagine you have a side-on camera view of a game of ping-pong, and you want to estimate the 3D coordinates of the ball at any given point in time using just this side view.
To achieve this, you could feed the 2D image coordinates and bounding box size of the ball from your side-on view into a Kalman filter, and generate your prediction matrix by recording a few objective 3D ball-position measurements from other camera angles.
Bounding box on its own wouldn’t give you an accurate measurement of the ball’s depth into the field of view (i.e. it’s a noisy signal), but you can use the bounding box to get a high-accuracy estimate of depth when you average over a few observations and correct for how the bounding box scales based on 2D coordinate position of the ball, as seen in the side-on camera view.
Yeah, there are apparently entire courses about how to use them. I've never actually used a Kalman filter, but they seem neat. It seems like some simpler examples would help. What are some...
Yeah, there are apparently entire courses about how to use them. I've never actually used a Kalman filter, but they seem neat.
It seems like some simpler examples would help. What are some practical special cases where a lot of the terms drop out? Does it make sense in one dimension?
I have to learn this for a project I'm working on and it seems... Vaguely ok?
Feels like it definitely could be more complex, and hopefully the accuracy is good enough
I think it would have been nice to have at least a few other examples of where KFs can be used. They tried to sell it by saying it can be used for all sorts of things, then just gave the same example you always see, robot/vehicle motion. That's like saying "a spoon can be used for so many different and unexpected things, for example, eating soup!" OP, got any suggestions?
Also, these are tough to implement without having good intuition on what these matrices look like in a realistic problem and what each part is doing. When I learned this in class, most of the questions were on this point. I guess the way they often teach linear algebra doesn't really prepare you for intuitively using it in real life problems.
Imagine you have a side-on camera view of a game of ping-pong, and you want to estimate the 3D coordinates of the ball at any given point in time using just this side view.
To achieve this, you could feed the 2D image coordinates and bounding box size of the ball from your side-on view into a Kalman filter, and generate your prediction matrix by recording a few objective 3D ball-position measurements from other camera angles.
Bounding box on its own wouldn’t give you an accurate measurement of the ball’s depth into the field of view (i.e. it’s a noisy signal), but you can use the bounding box to get a high-accuracy estimate of depth when you average over a few observations and correct for how the bounding box scales based on 2D coordinate position of the ball, as seen in the side-on camera view.
Yeah, there are apparently entire courses about how to use them. I've never actually used a Kalman filter, but they seem neat.
It seems like some simpler examples would help. What are some practical special cases where a lot of the terms drop out? Does it make sense in one dimension?