Name: open-traffic-poc-data-demo
Owner: Mapzen
Description: A day in the life on the roads of Manila, using data from the Open Traffic prototype
Created: 2016-11-18 17:51:57.0
Updated: 2018-01-05 10:33:03.0
Pushed: 2017-04-12 15:41:41.0
Homepage: https://mapzen.github.io/open-traffic-poc-data-demo/
Size: 46239
Language: JavaScript
GitHub Committers
| User | Most Recent Commit | # Commits |
|---|
Other Committers
| User | Most Recent Commit | # Commits |
|---|
data/ folderscript/process_export.py and edit EXTRACT_NAME valuescript/process_export.pyThis script generates a (data/opentraffic.png) and GeoJSON (data/opentraffic.json). The GeoJSON contain the geometry (A FeatureColection of LineString) which propertie.id match the pixel row of the PNG image.
Because there is a lot of geometries in Manila (arround 38563), I wrapp them in columns.
Edge Id,Date Start,Date End,Monday,Tuesday,Wednesday,Thursday,Friday,Saturday,Sunday,Time Start,Time End,...
1000001163,05/25/2015,10/02/2016,1,0,0,0,0,0,0,10:00,010:59
To have a consistent timestamp I'm taking the Date Start and offset it by that day of the week
# Calculate the timestamp from the "Date Start" ...
timestamp = datetime.strptime(row[1]+' '+row[10],'%m/%d/%Y %H:%M')
# ... plus adding the offset of days from Wednesday (note: this will not work probably for other extracts)
time_offset = 5*int(row[3]) + 6*int(row[4]) + 0*int(row[5]) + 1*int(row[6]) + 2*int(row[7]) + 3*int(row[8]) + 4*int(row[9])
timestamp = timestamp + relativedelta(days=time_offset)
In this case the week start on a Wednesday so I'm adjusting the offsets for the days manualy… which is not a great solution.
Time Start and Time End have weird values
For example: ...,010:59,... or ...,0-1:00,....
On first case: I take out the extra 0
For the second case: seems that the export don't have a 23:00 sample… so I assume that goes from 23:00 of the previous day.
I work around this by doing:time = row[10].split(':')
if len(time[0]) == 3:
time[0] = time[0][1:]
timestamp = datetime.strptime(row[1]+' '+str(1+int(time[0]))+':'+str(1+int(time[1])),'%m/%d/%Y %H:%M')
The big GeoJSON takes a long time to load. It's possible to condense all the data into a .mbtile using tippercanoe and then export in indivual tiles thanks to this script https://github.com/iandees/mbtiles2dir by @iandees (Ian Dees). Once both are installed do:
ecanoe -S 10 -o opentraffic.mbtiles opentraffic.json
on mbtiles2dir.py opentraffic.mbtiles --flipy opentraffic
-d -r -S .pbf *
. -type f -exec mv '{}' '{}'.pbf \;
I'm loading both the PNG image (data/opentraffic.png) and GeoJSON (data/opentraffic.json) to the Tangram YAML scene file scene.yaml, where I colorize each segment of the geometry according to the geometry ID (that match the row on the image)
rs:
opentraffic:
data: { source: opentraffic }
draw:
opentraffic:
order: 10001
width: [[10,2px],[15,5px],[20,5m]]
color: |
function () {
return [ ( Math.floor(feature.id/255))/255, (feature.id%255)/255, 0 ];
}
Then on the Shader style:
t getID (vec2 encodedID) {
return (encodedID.x*65025.+encodedID.y*255.)+.5;
getUV (float id, float x_offset) {
float y = mod(id,u_param.y);
float x = floor(id/u_param.y)*COL_WIDTH;
return vec2(x+x_offset, y);
data_now = getColor(getUV(id,i_now));
data_next = getColor(getUV(id,next));
data_interpol = mix(data_now, data_next, f_now);
st = v_texcoord.xy+vec2(.5,0.);
-= u_time*5.*data_now.r;
r.rgb = mix(vec3(1.000,0.068,0.378),vec3(0.000,1.000,0.818),smoothstep(0.,.3,data_interpol.r));
r.a *= aastep(zoom(),fract(st.y+abs(st.x*.5-.5)));
Why Chrevrones and not cars? We only have the average of time on a segment… not so much information about how many cars. The fact that a road goes fast or slow don't indicate at this point how much traffic it have but at what speed people drive them. That's why I'm thinking it as flows, currents or wind fields.