Abstract:
The rapid development and proliferation of global positioning system (GPS)-enabled
systems and devices have led to a signi cant increase in the availability of transport
data, more speci cally GPS trajectories, that can be used in researching vehicle activities.
In order to save data storage- and handling costs many vehicle tracking systems
only store low-frequency trajectories for vehicles. A number of existing methods used
to map GPS trajectories to a digital road network were analysed and such an algorithm
was implemented in Multi-Agent Transport Simulation (MATSim), an open
source collaborative simulation package for Java. The map-matching algorithm was
tested on a simple grid network and a real and extensive network of the City of Cape
Town, South Africa. Experimentation showed the network size has the biggest in-
uence on algorithm execution time and that a network must be reduced to include
only the links that the vehicle most likely traversed. The algorithm is not suited for
trajectories with sampling rates less than 5 seconds as it can result in unrealistic paths
chosen, but it manages to obtain accuracies of around 80% up until sampling sizes of
around 50 seconds whereafter the accuracy decreases. Further experimentation also
revealed optimal algorithm parameters for matching trajectories on the Cape Town
network. The use case for the implementation was to infer basic vehicle travel information,
such as route travelled and speed of travel, for municipal waste collection
vehicles in the City of Cape Town, South Africa.
