Details about the computation of the pdf

[ThoChat]

Can you please check if the new details make sens for you and if you see any spelling mistakes.
Cheers!

[mirakuepper]

I have one additional question about randomization: Are the future trajectory points of a given person taken into account as well or only the position of all other people?

[ThoChat]

I have one additional question about randomization: Are the future trajectory points of a given person taken into account as well or only the position of all other people?

I'll not sure I understood the question correctly. Are you talking about the randomization process or about the computation of $P_{Ni}$?

[mirakuepper]

Ok, I'll try to explain my question:
For P(r) we calculate the distance each pedestrian has to all other people in the same frame. This we do for all persons and all frames and then we get the probability P(r).

But what do we do for PNi?
Do we ignore the frames and then see how the distances for all people are distributed?
Or do we drop the frame-information and then pick positions at random (but how many do we pick?)?
Or do we calculate this per person as well? Like taking one persons position and then calculate the distance to the randomised positions (but if we do this, a person could be compared to his own future/past path)?

This somehow really confuses me...

[ThoChat]

Ok, I'll try to explain my question: For P(r) we calculate the distance each pedestrian has to all other people in the same frame. This we do for all persons and all frames and then we get the probability P(r).

But what do we do for PNi? Do we ignore the frames and then see how the distances for all people are distributed? Or do we drop the frame-information and then pick positions at random (but how many do we pick?)? Or do we calculate this per person as well? Like taking one persons position and then calculate the distance to the randomised positions (but if we do this, a person could be compared to his own future/past path)?

This somehow really confuses me...

I get it now. : )
Thanks a lot for detailing your question.
The answer is the second option you mentioned : We drop the frame-information and then pick positions at random. The number of positions picked is then also random. The random process should ensure a uniform distribution of positions for each frame. However, to smooth any noise that this method may induce, it's better to use a higher `randomisation_stacking' number for the computation of the function. This is shown in the "Parameters of the PDF" section in the User guide.

[mirakuepper]

Thanks a lot! Can you maybe add the information that the number of picked positions is also randomised to the guide? From "randomisation-stacking" I thought we did the random picking several times but not that we change the number of positions picked.
But I think now I understand what is done here :D