A simple approach to smooth a GPX track without loosing too much accuracy.
usage: GPX smoother [<flags>] <file>
Soe for example: gpx-smoother -d -s 10 -w 6 2019-09-19_12-12-16_new.gpx
Detailed information with flag -h, --help.
| parameter | default | description |
|---|---|---|
-w, --weight |
3.0 |
Specifies how strong the smoothing should happen. Larger numbers result in a more precise track, lower numbers in a smoother one |
-s, --size |
6 |
Specifies how much surrounding point of each GPX-point should be considered. Larger numbers result in a more precise track, lower numbers in a smoother one |
-d, --debug |
(not set) | Verbose mode, showing additional debug information |
-h, --help |
(not set) | Shows information about the usage |
It depends on the density of the track points.
For tracks with points every 5 meter the default parameters (-w 3 -s 6) are fine. For dense tracks a larger size-value (e.g. -s 10) might be better.
Currently this takes each point of the track and gets n neighbors around it. Then a weighted average of their locations is calculated: Points near by are weighted more than points far away.
// Basically: How strong is the weight. This is later an exponent.
weightFactor := 3.0
// for each point:
// for each neighbor of the current point
// Weight look like: 1 2 3 [4] 3 2 1 (where [4] is the point currently looking at)
// This is used as basis for "weight^weightFactor"
weight := stepsBackwards/2 - int(math.Abs(float64(j-i))) + 1
// sum up the coordinates
latSum += points[j].Lat * math.Pow(float64(weight), weightFactor)
lonSum += points[j].Lon * math.Pow(float64(weight), weightFactor)
// To later get the correct location, we need to know how many locations we used
weightSum += math.Pow(float64(weight), weightFactor)
// Get the average
newTrkpt[i-stepsBackwards/2] = Trkpt{
Lat: latSum / weightSum,
Lon: lonSum / weightSum,
}