search

datamapio / posmo_one_testing

Public Testing Repo
0 stars 0 forks source link

Walking distances are off #31

Open thohug opened 3 years ago

thohug commented 3 years ago

Setting Zurich, Switzerland, Android

wifi-fence: 1 do_not_track: 0 wifi_over_gps: 1 data_over_wifi_only: 1

Issue Walking distances are wrong and differ strongly from reality (probably due to GPS inaccuracies).

9mins for 1km is quite fast: grafik

Distance from google maps is about half of the distance measured: grafik

rogerfischer commented 3 years ago

This is about the distance, right? I think that wrong distance is definitely due to GPS noise. Only a graph based solution can correct this.

Screen Shot 2020-12-18 at 9 30 53 PM

](url)

thohug commented 3 years ago

@rogerfischer or smoothing the line? I feel like this could be quite an issue for statistical purposes, as these lengths need to be reliable...

djr4 commented 3 years ago

It would be possible to apply kalman filtering before doing distance calculation, line would be smoother thus more accurate (At least that is my intuition)

On Sun, 20 Dec 2020 at 13:17, thohug notifications@github.com wrote:

@rogerfischer https://github.com/rogerfischer or smoothing the line? I feel like this could be quite an issue for statistical purposes, as these lengths need to be reliable...

— You are receiving this because you are subscribed to this thread. Reply to this email directly, view it on GitHub https://github.com/datamapio/posmo_one_testing/issues/31#issuecomment-748600337, or unsubscribe https://github.com/notifications/unsubscribe-auth/ABQORJF5B57MUFBJARZZRXLSVXTO7ANCNFSM4VAXWZ7Q .

rogerfischer commented 3 years ago

The problem with smoothing is that you might have a straighter line, but it might be over buildings. I think the only way to make this perfect is to use a graph (=> OSMnx).

Next week I will start with the train layer and then we can work from there.

PS: @thohug How many walking examples do you have like this? My walking is - if off - in an acceptable range for now. 

  PS: The best would be to use OSM to show the path you walked really and compare the distances.
thohug commented 3 years ago

@rogerfischer Ich versuche ein Bisschen darauf zu schauen. Ist mir aber schon ein paar Mal aufgefallen.

rogerfischer commented 3 years ago

@thohug Auf https://www.openstreetmap.org/directions#map=17/47.41088/8.54047&layers=T kann man die Wege ausmessen. Da können wir vergleichen, wie stark wir davon abweichen.

Screen Shot 2020-12-21 at 2 53 48 PM

PS: Im grossen Ganzen (ausser beim Zug) sollten die Abweichungen nicht zu gross sein. Und sobald wir mit Graphen schaffen, wird es dann noch genauer. Aber das dauert noch ein wenig.

djr4 commented 3 years ago

Note on this: We have barefoot map matcher but unless I only pick start and end point we have some strange paths. Like intersection forward and back if you remember.

rogerfischer commented 3 years ago

@djr4 Yes, the map matching algos are not ideal. There are other ways to obtain better results. E.g. multiple paths for a given segment.

djr4 commented 3 years ago

By the way I'm testing right now some corrections that will make this much better with raw gps data, it does feel we are off by 150m or so depending on accumulation of points near static point. (noise)