I have been provided a dataset that was generated using UWB indoor localization of individuals and I notice that each timestamp has 2 associated different coordinates. This is rather general application oriented doubt and I would appreciate any guidance around how to decide which point to select for application purpose?
A blind guess is x2 tags or two individuals ?
UWB is intended for positioning in 3D or 2D and the result depend on the reference used by the triangulation algorithm.
You did not specify which algorithm or fimware is being used in your case, they’re all different and can decide arbitrarily which coordinates system to use.
An option you have is to identify that by plotting both coordinates and see if they have an offset relative to each other or any other correlation, e.g. if both are moving within the same space by chance. At worst asking the database provider is the most reliable way.
Thanks Wassfila for your kind input. Apologies for I am hardware/firmware agnostic and just starting with UWB on software/application side. To clarify further, I am looking at (by plotting) individual Id/badge at a time and to my surprise every timestamp has 2 associated coordinates. The points do move together from POI to POI (but also spatially differ within respective POIs and sometimes also crosses POI, while staying not too far from each other). I am wondering about the reason/purpose behind this. Also, how to handle such cases?
Could it be because UWB hardware setup configuration was such that each id was being tracked twice for redundancy and correction purposes?..well, looks like it and I am thinking of using their mean.
No issues, I think judging from what you see seams reasonable.
It could be result of 2 tags moved together or two algorithms maybe based on triangulating different anchors.
Usually in such cases to evaluate the accuracy or compare algorithms, you need a ground truth sample. A special measure made knowing time and pos measured with ruler mm precision.
Without that, you could create a filter model that matches your systems dynamics (e.g. person) and check which measure deviate less from the filtered measure.
thanks a lot Wassfila, what you’ve shared makes total sense. Can you possibly also recommend any robust Stop-Move algorithm/open-source?.. trying to make visual sense of lengthy spiky trajectories.
hmmm, to be honest, it’s a wide topic, it can go from the simple Schmitt trigger (on higher than off) to avoid instability, up to a neural net that maps a complex non linear function with multiple outliers.
This can have a statistic perspective as well as a systems dynamics one depending on the needed quality and effort you’re willing to spend, for that multiple open source libraries exist. Usually, tweaking the uwb algorithms parameters can deeply impact the quality of your measures depending on the use case.
For info, everything I develop in the context of uwb I’ll be sharing it open source but I’m only getting started with uwb R&D.
I started a review of existing open source sw related to uwb and the Decawave modules here
I also provide free support for projects closely related to the ones I’m working on, if you want you can get in touch.
Cool, you have so much going on …fairly rich repo. I have only recently started on the uwb sw application side and can use some guidance. What is the best way to reach out to you? thanks!
For brainstorming about open source projects not directly related to Decawave support, we could use this I have put in place for home automation in general