Summarize GPS information#
This functionality of tracklib reduce the full dataset of GPS traces into a regular grid of summarized features. In each cell, n aggregated features (such as mean and standard deviation of speeds, number of traces, most frequent bearing …) are computed to produce a set of raster maps, which may be seen as a single image with n channels.
These process has been used in works presented in (1).
Figure 1 : Workflow presented in (1) to produce raster images (5 meter-resolution grid)
Reference:(1) - Y. Méneroux and al. Convolutional Neural Network for Traffic Signal Inference based on GPS Traces. Proceedings of Spatial big data and machine learning in GIScience Workshop’18, August 28-31, 2018, Melbourne, Australia. GIScience.
Import de la librairie tracklib#
Il faut importer tracklib si la librairie n’est pas dans le python path
[1]:
# -*- coding: utf-8 -*-
import matplotlib.pyplot as plt
import os.path
import sys
#-------------------------------------------------------
# Import de tracklib
module_path = os.path.abspath(os.path.join('../../..'))
if module_path not in sys.path:
sys.path.append(module_path)
Chargement du jeu de traces#
The dataset comprises a set of 11862 GPS traces, located in Mitaka city (16 km2), suburbs of Tokyo (Japan).
[2]:
from tracklib.io.TrackReader import TrackReader as reader
from tracklib.core.ObsTime import ObsTime
PATH = '/home/marie-dominique/PROJET/FINI/MITAKA/DATA/mitaka/complet'
ObsTime.setReadFormat("4Y-2M-2D 2h:2m:2s")
collection = reader.readFromCsv(path=PATH, id_E=2, id_N=3, id_T=1)
print ('Number of tracks: ' + str(collection.size()))
Number of tracks: 11872
Sélection des traces sur une zone restreinte de type Rectangle#
[3]:
import matplotlib.pyplot as plt
from tracklib.core.TrackCollection import TrackCollection
from tracklib.core.Track import Track
from tracklib.core.ObsCoords import ENUCoords
import tracklib.algo.Geometrics as Geometrics
from tracklib.algo.Selection import Constraint
from tracklib.algo.Selection import TYPE_CUT_AND_SELECT, MODE_INSIDE
Xmin = 370250
Xmax = 371250
Ymin = 3949050
Ymax = 3950050
ll = ENUCoords(Xmin, Ymin)
ur = ENUCoords(Xmax, Ymax)
bbox = Geometrics.Rectangle(ll, ur)
constraintBBox = Constraint(shape = bbox, mode = MODE_INSIDE, type=TYPE_CUT_AND_SELECT)
selections = constraintBBox.select(collection)
print ('Number of tracks: ' + str(selections.size()))
for track in selections:
if track.size() < 10:
selections.removeTrack(track)
selections.plot('k')
plt.xlim((Xmin, Xmax))
plt.ylim((Ymin, Ymax))
plt.show()
print ('Number of tracks: ' + str(selections.size()))
collection = selections
Number of tracks: 3850

Number of tracks: 3804
Raster de la moyenne des vitesses et de l’orientation de la plus fréquente par cellule#
Les AF seront calculées si elles n’ont pas encore été créées
Les scripts des deux AF qu’on prend existent déjà dans la librairie, sinon il aurait fallu les écrire
[4]:
from tracklib.algo import (Analytics, Summarising)
# On crée l'AF speed
collection.addAnalyticalFeature(Analytics.speed)
af_algos = [Analytics.speed, Analytics.orientation]
cell_operators = [Summarising.co_avg, Summarising.co_dominant]
marge = 0
resolution = (5, 5)
raster = Summarising.summarize(collection, af_algos, cell_operators, resolution, marge)
Conversion des indicateurs en 8-bit grayscale#
Feature values are converted on a 8-bit grayscale.
[5]:
# the mean of speeds is distributed between 16 and 255
band1 = raster.getRasterBand('speed#co_avg')
for i in range(band1.nrow):
for j in range(band1.ncol):
val = band1.grid[i][j]
if val != band1.getNoDataValue():
if val > 25:
band1.grid[i][j] = 10
else:
band1.grid[i][j] = 100 - val/25*90
# The dominance of orientations is discretized in 8 eight values
band2 = raster.getRasterBand('orientation#co_dominant')
for i in range(band2.nrow):
for j in range(band2.ncol):
val = band2.grid[i][j]
if val != band2.getNoDataValue():
if val == 1:
band2.grid[i][j] = 156
elif val == 2:
band2.grid[i][j] = 218
elif val == 3:
band2.grid[i][j] = 249
elif val == 4:
band2.grid[i][j] = 187
elif val == 5:
band2.grid[i][j] = 125
elif val == 6:
band2.grid[i][j] = 63
elif val == 7:
band2.grid[i][j] = 32
elif val == 8:
band2.grid[i][j] = 94
else:
band2.grid[i][j] = 0
Affichage#
[6]:
fig = plt.figure(figsize=(16, 8))
ax1 = fig.add_subplot(121)
band1.plotAsImage(axe=ax1, figure=fig, novaluecolor='teal')
ax2 = fig.add_subplot(122)
band2.plotAsImage(ax2, figure=fig, novaluecolor='teal')
