[OSM-talk] Determining direction of travel and segment connectivity from GPS traces

[Nick Hill]

It occurred to me yesterday when I described in the pub on IOW that we 
could infer routing information from the GPS traces by adding another layer.

I visualised the layer as being a grid of hexagonal magnetic domains 
with a series of properties.

1) Vector
2) Bi-directionality
3) Field Strength

An improvement, but with a more complex magnetising and reading 
algorithm could have

1) n number of planar vectors
2) Each vector having a given field strength.

Or

Adjacent hexagon linkage value where the linkage value is only 
incremented in direction of travel.

If we are thinking of creating a system which can determine valid 
routes, we will need to determine valid routes for specific modes of 
transport. For example, separating routability information for:

Car
Train
Bus
Walking
Cycling
Boating

into sepearate routability layers for each transport mode.

Ideas how to determine traces for different modes of transport:

[Upper quartile speed]
Walking <5mph
Driving >20mph
Ferry <20mph
Cycling 8-16mph
Bus <20mph

[Vector change]
Boating : Low all others, high

[Velocity variance]
Ferry low
Driving medium
Bus High

[Velocity Variance based on gradient]
Cycling: Very high positive velocity varience based on rate of elevation 
drop
Bus: medium
car low
Boating: Very small gradient so N/A

[Upper quartile rate of acceleration]
Boat Very low
Cycle Low
Bus medium
Car high
Walking Low

[Velocity varience when changing vector]
Car Small
Boat N/A
Cycle Small
Bus high


Method of application

Create a score for whole track with weighted average of 20%
Divide track in half. Apply same analysis to each half. Give weighted 
average of 20%
Sub-divide again into 4. Apply a 20% weighting for each quarter
then eigth, sixteenth, 1/32nd.
Each 1/32nd segment has 5 scores. For the whole track then each 
sub-division down to the 32nd. Summing each score should average noise 
caused by different travelling environments on the same type of transport.