Correcting Imperfect Sample Rates in Trip Matrices Derived from Automatic Number Plate Recognition Cameras – a Simulation Approach



Correcting Imperfect Sample Rates in Trip Matrices Derived from Automatic Number Plate Recognition Cameras – a Simulation Approach

Authors

Ben Mackley, Mott MacDonald, Tom Van Vuren, Mott MacDonald

Description

This paper describes the development of a method for correcting the impact of variable ANPR sample rates on building OD matrices. A simulation approach coded in Python is used to evaluate the method using real-life ANPR data.

Abstract

In today’s world of big and passive data collection, transport modellers and planners are continually replacing traditional methods with evolving technologies such as those based on GPS and mobile phone data. Whilst these data sources are enabling the development of OD matrices for higher level strategic networks, obtaining detailed traffic behaviour information to refine such datasets (spatially but also with respect to mode and purpose) requires more traditional methods. This brings inherent risk with their low sample rates and laborious and error-prone collection processes and hence there is a need to find a data source that can provide precise OD information, without compromising on sample size.

Automatic Number Plate Recognition (ANPR) offers a method of creating detailed OD matrices with the ability to sample the entire network population. The detail it provides in terms of origin and destination zones and routing is dependent on how the network of cameras is structured; a higher density produces finer granularity. Through analysis of number plates, trips can be disaggregated by vehicle type and by emissions class – the latter is of particular interest for air quality studies. Further to this, since number plates are tracked at known locations and times during the survey, journey time and routing information can be extracted and used in calibration and for other planning uses.

In this paper we focus on how these OD matrices are affected by ANPR sample rates, the measure of how successful an ANPR camera has been in capturing passing number plates. The ability to accurately track a number plate across a network relies on all sample rates being close to 100%. As stated in (Chen et al., 2015), ideal conditions can bring sample rates well above 85%. However we have observed that external conditions such as adverse weather conditions and the positioning of a camera can lead to lower sample rates.

To explore how significant this issue is, we have coded a simulation approach in Python that takes a dataset that represents reality (in the format of ANPR data) and simulates the impact of variable ANPR sample rates. We then study how the resulting OD matrices compare before and after sampling and use this to design and test corrective measures that are applied to ANPR data. In this paper we present a correction approach and apply it to trip matrices built across 3 configurations of ANPR cameras, from a real-life study in the UK, using the simulation approach to demonstrate its effectiveness.

Through this paper, we aim to increase confidence in the use of ANPR as a data source for building OD matrices. Methods established in (Castillo et al., 2012), on design of ANPR networks, describe how best to structure an ANPR network to make most efficient use of resources. We hope that alongside existing work like this, our paper can help ANPR to be used to its full potential in transport studies not just in the UK, but across Europe. It is also hoped that the simulation approach may encourage similar efforts to quantify the impact of sampling errors in other survey types, such as roadside interviews, mobile phone data or GPS-based data.

Chen, Tse-Shih, Ming-Fen Lin, Tzi-Cker Chieuh, Cheng-Hsin Chang, and Wei-Heng Tai. "An Intelligent Surveillance Video Analysis Service in Cloud Environment." 2015 International Carnahan Conference on Security Technology (ICCST) (2015): n. pag. Web.
Castillo, Enrique, Ana Rivas, Pilar Jiménez, and José Marí¬a Menéndez. "Observability in Traffic Networks. Plate Scanning Added by Counting Information." Transportation 39.6 (2012): 1301-333. Web.

Publisher

Association for European Transport