Origin ­destination analysis on the London Orbital Automated Number Plate Recognition network



Origin ­destination analysis on the London Orbital Automated Number Plate Recognition network

Authors

C Fox, University of Sheffield, UK; P Billington, Telematics Technology, UK; D Paulo, Mouchel, UK; C Cooper, Highways Agency, UK

Description

Abstract

The UK Highways Agency (HA) has an interest in understanding the actual journeys taken by
drivers using the London orbital motorway (M25) in order to assist with planning and management
of strategic routes. It is thought that the majority of traffic on Public Service Agreement (PSA)
Routes, rather than travelling end to end is joining and leaving mid­route, possibly travelling
between only one or two junctions.
Motorway Incident Detection and Signalling (MIDAS) data is available containing count, speed and
occupancy information, providing an understanding of stresses on individual links and the turning
movements of traffic at specific junctions. However, this data does not readily provide useful
estimates of where vehicles start and end their journeys.
The HA's goal is the construction of an Origin­Destination (O­D) matrix, in the form of a look­up
table having all possible origins along one axis and all possible destinations along the other, each
cell in the matrix giving the number or relative proportion of vehicles making the journey between
the relevant orthogonal origin/destination pair.
The present ANPR infrastructure (National Traffic Control Center, NTCC) on the M25 and its
arterials was designed for other purposes and in particular does not provide full lane coverage.
Instead, single or pairs of cameras are positioned at sites covering subsets of the lanes. The
proportion of traffic detected by these partial lanes varies according to the flow of traffic. For
example, a camera in lane 1 of a three­lane motorway will detect most of the traffic during low
flows, but only a third of the traffic when the motorway is saturated. Conversely, a pair of cameras
on lanes 2 and 3 will detect a low proportion of traffic during low flows but a higher portion nearing
saturation. A single camera in lane 2 may show a more complex flow/detection relationship.
The situation is further complicated by the hashing scheme used by the existing infrastructure. The
existence cameras use a hashing scheme which represents each plate by an 18­bit tag. An 18 bit
value provides a maximum of 262144 possible tags to represent approximately 30 million vehicles
in the UK. Thus matches between tags at the origin and destination of a route must be processed to
removed spurious matches due to hashing clashes, as well as to adjust for the flow­affected
detection proportion.
In this study, we recorded one day of calibration data from 12 routes of different lane configurations on the M25, using higher quality cameras (Telematics Technology, Sheffield, UK) than NTCC cameras, having a higher detection rate and reporting raw plate numbers rather than hashes. The
calibration cameras were placed to cover all lanes of the calibration routes, and allow us to analyse
how the proportion of traffic in each lane varies as a function of total flow as measured by MIDAS
data. These multipliers are used in conjunction with a novel tracking algorithm to obtain estimates
of origin­destination route profiles for a variety of day types.
The tracking algorithm maintains an adaptive window of journey durations between origin and
destination, used to filter out spurious matches due to redundant number plate hash codes. Thus
hard classification of genuine/spurious matches is performed concurrently with updating beliefs
about the distribution of journey times throughout the day ? a form of expectation­maximisation
algorithm (A.P.Dempster, N.M. Laird, and D.B. Rubin. Maximum likelihood from incomplete data
via the EM algorithm. Journal of the Royal Statistical Society. B, 39(1):1?38, 1977.) A further
analysis detects if the adaptive tracking becomes lost, and iteratively adjusts the window parameters
to regain it.
We present quantitative results showing how the detection rate from various restricted lane camera
configurations varies as a function of MIDAS traffic flow. These functions show significant
nonlinearities and are used to estimate traffic on particular routes from restricted NTCC camera
data.
The results of the calibration study were applied to estimating the number and duration times of
journeys made on 108 particular routes on the M25 and its arterials, over an eight week period in
2009. (Our algorithm requires about 3 hours to process the 8Gb of raw data for this period on an
Intel Centrino Duo.) We have constructed a GUI which allows users to select routes and display
their durations as a function of time of day. The tool enables a greater understanding of origin­
destination movements and the operational implications of the real routes taken by traffic around the M25 and on relevant arterials.
We have submitted the draft full paper along with this abstract to aid with reviewing.

Publisher

Association for European Transport