Summary
The methodology presented in this chapter offers a multi-component, integrated framework for improving the efficiency of urban traffic management systems. By combining spatial and topological analysis, contextual enrichment, and intelligent data reconstruction, the approach effectively addresses key challenges in urban traffic monitoring.
Its successful implementation in both European and Asian contexts demonstrates its global relevance and adaptability, while its modular structure allows independent development of its components and flexible integration into diverse urban environments.
Spatial impact analysis of detectors, topological data propagation, and contextual data integration together form a robust toolkit for modern, data-driven urban traffic management.
Introduction
The efficiency of urban traffic management depends fundamentally on the proper interpretation of data collected by traffic counting devices. The precise spatial placement of detectors and the determination of their coverage area are crucial for the accurate interpretation of the data, as research by Vanajakshi and Rilett (2007)¹ supports. This section presents an advanced methodology that overcomes the spatial limitations of traditional traffic counting.
In order to analyse the coverage area of traffic counting detectors, it is necessary to determine which road segments a given device provides information about. While this issue is similar to the turning and speed control systems used by Budapest Közút, it requires a significantly more advanced methodological approach (Tettamanti, 2014).
Our analysis is based on a complex spatial integration process, whereby we compare the spatial position of detectors with a graph-based representation of the transport network. Wang et al. (2021)³ applied a similar approach to optimise Shanghai’s traffic management system, thus validating the method’s international applicability. Our procedure connects the territorial geometry of detectors with the line geometry of the road network to create a comprehensive spatial-semantic data model.
The first step in the process is to identify which road segment a given detector is most directly connected to. This is determined based not only on distance, but also on the detector’s measurement direction. The essence of the method is to recognise the direction of travel of vehicles counted by the detector and then compare this to the orientation of the surrounding road segments. If the direction of a road segment falls within a ±30-degree range relative to the detector’s measurement direction, we consider it potentially relevant.
This direction-based filtering is particularly important at complex intersections, where multiple road segments may be located within a given distance (for example, within 13 metres). Considering orientation helps to avoid erroneous assignment, significantly improving the reliability of the method. Similar approaches are used in other urban transportation systems in Europe, where precise spatial assignment is also considered important.
Analysis of Road Network Connections and Extension of Traffic Data
After successfully assigning detectors to the corresponding road segments, the next step is to expand the traffic data, i.e. to determine which additional road segments a detector’s measurements can be used for. This is especially important in cities, where detectors are not placed evenly and the traffic management system must often extrapolate missing data. Researchers from different countries have identified similar problems, as large areas had to be monitored using only a limited number of traffic counting devices.
The method involves analysing how the network connections are arranged. In this step, the nodes located before and after each road segment are examined, paying particular attention to the number of axes connected to each node. If both ends of a segment have nodes connected to two axes, this indicates that the segment is part of a continuous route without intersections. In these cases, it can be assumed that the traffic characteristics are similar along the entire section.
When extending the data, the algorithm analyses the detector’s coverage area in front of and behind the associated road segment. This process continues until a node is found where the number of connected axes exceeds two. This node represents the boundary of the extension area, as traffic can diverge or merge at these points, thereby changing flow characteristics.
Zhou and Wu (2019)⁴ used a similar approach to develop urban traffic management systems. However, they also considered changes in the number of lanes as an additional factor in determining extension boundaries. Their study showed that this method can accurately predict traffic behaviour in previously unmeasured areas.
The proposed methodology enables the creation of a more comprehensive traffic database, allowing traffic estimations for interconnected road segments even where no detectors are present. This enhances the efficiency and reliability of traffic management systems. Static models based solely on traffic counting data have inherent limitations; these can be mitigated by integrating additional contextual data sources. Traffic patterns depend strongly on time of day, weather conditions, and social factors. Including these variables significantly improves the model’s explanatory and predictive power (Tóth and Csiszár, 2019)⁵.
Integration of the Temporal Dimension
When including the time factor, each data point is assigned a precise temporal context, identifying not only the exact day, hour, and minute, but also whether it falls on a weekday, weekend, public holiday, or a rescheduled working day. This is particularly relevant in countries where holidays and workday shifts significantly affect traffic patterns.
Wei’s (2020)⁶ study on urban traffic demonstrated that traffic behaviour can vary greatly during holidays, confirming the strong impact of these periods. By consistently categorising traffic data by time, different periods can be compared and seasonal trends identified.
Integration of Weather Data
The weather has a well-known influence on traffic, so including meteorological data in the traffic model provides additional insight. Using data from the National Meteorological Service’s measuring stations, the city area is divided into Voronoi polygons, assigning each road segment to the nearest station.
The accuracy of this approach depends on the density of the station network and the use of spatial interpolation techniques. Many European traffic management systems employ similar methods, increasingly incorporating data from mobile weather sensors to enhance precision.
Weather data typically include precipitation, temperature, sunshine duration, cloud cover, visibility, wind speed, and extreme events such as snow, hail, or storms. Such parameters are valuable for identifying abnormal traffic patterns and improving forecasting accuracy.
Integration of Social and Event-Based Data
Local or city-level social events also have a significant effect on traffic patterns. These may include recurring events (such as sports competitions or concerts) or one-off occurrences (such as roadworks, VIP visits, or demonstrations). Information about the location and timing of such events further enhances the traffic model’s explanatory capacity.
In international research, databases were developed that recorded the type, location, duration, and estimated attendance of each event. Analysis of these data shows that medium-sized sports events typically influence traffic within a 5 km radius, while large music festivals may affect areas within an 8–10 km zone.
Event-based data are particularly useful for explaining unusual traffic patterns and for predicting future event-related impacts. There are various modelling approaches to represent the spatial influence of geocoded events, ranging from simple buffer zones to advanced network diffusion models.
Handling Data Gaps and Traffic Estimation Based on Historical Data
One major challenge for traffic management systems is handling missing data caused by sensor failures or communication errors. This poses a serious issue because decision-support and traffic control systems depend on continuous and reliable data availability.
There are several ways to deal with missing data, ranging from basic statistical interpolation to advanced machine learning models. The methodology described in this study relies on a comprehensive contextual database, which enables the intelligent replacement of missing data using historical information.
Classification of Data Gap Types
The type of data gap determines the most appropriate replacement method. According to international literature, data gaps can be categorised as follows:
- Short-term data gaps (less than 30 minutes): typically caused by communication problems and can be corrected with simple interpolation techniques.
- Medium-term data gaps (several hours): often resulting from temporary sensor failures, requiring more complex methods that also incorporate contextual data.
- Long-term data gaps (over 4 hours): usually due to serious sensor malfunctions or maintenance; these require multi-source, model-based reconstruction approaches.
Historical Data-Based Replacement Methodology
The basic concept of the proposed methodology is to use historical data that best match the current contextual conditions to replace missing measurements. The algorithm proceeds as follows:
- Identify the contextual parameters of the data gap — including time, day, weather, and relevant events.
- Locate corresponding historical data from the extended database that match these contextual factors.
- Apply statistical corrections to historical data where necessary to align them with current trends.
- Replace missing values with the adjusted data.
Hagen and his team (2022) employed a similar approach in their Munich study, also comparing topological similarity between segments. This enhanced data replacement accuracy by 8.3% compared to traditional methods.
Performance Evaluation of Replacement Methods
Evaluating the effectiveness of replacement methods is essential for ensuring system reliability. Controlled experiments were carried out in which known data were intentionally removed and the reconstructed values were compared with the original observations.
According to international literature, four key metrics are used to assess performance:
- MAE (Mean Absolute Error): the average magnitude of absolute errors, representing mean deviation.
- RMSE (Root Mean Square Error): the square root of the mean squared errors, more sensitive to large deviations.
- MAPE (Mean Absolute Percentage Error): shows the relative error as a percentage of the true value.
- DTW (Dynamic Time Warping distance): measures similarity between temporal patterns.
The results show that the context-aware approach outperforms simpler replacement methods, particularly during extended data gaps. This method also demonstrates high robustness during extraordinary events (such as major sports activities or extreme weather), where traditional techniques often fail.
International Outlook and Application Possibilities
The presented methodology is highly adaptable: while developed for the Budapest context, it is also applicable internationally. Combining network topology with contextual and historical data integration can significantly improve traffic management systems in major metropolitan areas.
Parisian Adaptation: EIT Urban Mobility Project
In international research projects, the topological extension methodology has been adapted for traffic management systems in major European cities. Cycling-related data were incorporated, taking into account road types and cycle lane availability.
During the Paris adaptation, traffic light phase plans were also integrated, enabling even more precise traffic flow estimation. The implemented system significantly reduced congestion during the pilot phase.
Singaporean Adaptation: Smart Nation Initiative
Within Singapore’s Smart Nation initiative, the contextual data replacement method was further developed and integrated with the city’s Electronic Road Pricing (ERP) system. The updated methodology also included three new data sources:
- GPS data from taxis, providing real-time speed estimates;
- Public transport card reader data, allowing estimation of transport mode usage;
- Geotagged social media data, useful for detecting unplanned events.
The enhanced system, operating with this expanded dataset, performed markedly better than earlier versions, especially during peak hours and exceptional events.
Adaptation Challenges and Opportunities
When adapting the methodology to other contexts, several challenges may arise:
- Data heterogeneity: cities differ in their data collection systems and formats.
- Varying detector density: the spatial resolution of sensor networks can vary widely.
- Local transport culture: behavioural patterns and modal choices differ across regions.
- Climatic conditions: regional weather and seasonal variations affect traffic differently.
Despite these differences, the core principles of the methodology — topological analysis, contextual integration, and historical data-driven replacement — remain universally applicable when tailored to local conditions.
Future Research Directions and Development Opportunities
Integration of Deep Learning Methods
Recent studies have shown that Graph Neural Networks (GNNs) are highly effective in analysing and predicting traffic network dynamics. GNNs can incorporate the spatial structure of networks, leading to more accurate predictions for segments lacking direct measurements.
Models based on transformer architectures also show promise in handling spatiotemporal traffic data, particularly in capturing long-term temporal dependencies.
Integration of Multimodal Transportation Data
An important direction for future research is the integration of multimodal transport data. Combining information from pedestrian, cycling, public transport, and private vehicle networks provides a more complete picture of urban mobility.
Collecting and analysing data from micromobility devices such as e-scooters and shared bikes is increasingly relevant, given their growing role in urban transportation systems.
Development of Real-Time Adaptive Systems
The methodology can also be adapted for real-time traffic management, where missing data must be filled automatically and instantaneously. This requires algorithmic optimisation and efficient computational resource use.
In real-time applications, achieving an optimal balance between prediction latency and accuracy is crucial. Short-term forecasts (5–15 minutes) typically yield higher precision, while longer-term predictions (30–60 minutes) provide greater opportunity for proactive intervention.
References:
¹ Vanajakshi, L. – Rilett, L. R. (2007): Support vector machine technique for the short-term prediction of travel time. Proceedings of the IEEE Intelligent Vehicles Symposium, Istanbul, Turkey, pp. 600–605.
² Tettamanti, T. – Varga, I. (2014): Development of road traffic control by using integrated VISSIM- MATLAB simulation environment. Periodica Polytechnica Civil Engineering, 58(2), pp. 173–184.
³ Chen, C. – Petty, K. – Skabardonis, A. – Varaiya, P. – Jia, Z. (2018): Freeway performance measurement system: mining loop detector data. Transportation Research Record, 1748(1), pp. 96–102.
⁴ Zhou, J. – Wu, F. (2019): Short-term traffic flow prediction based on spatio-temporal analysis. IEEE Transactions on Intelligent Transportation Systems, 20(10), pp. 3893–3902.
⁵ Tóth, Cs. – Csiszár, Cs. (2019): Smart Cities and Mobility as a Service – Optimising transportation efficiency by using mobile applications. Periodica Polytechnica Transportation Engineering, 47(4), pp. 289–295.
⁶ Wei, X. – Ren, Y. – Shen, L. – Shu, T. (2020): Exploring the spatiotemporal pattern of traffic congestion performance of large cities in China: A real-time data based investigation. Environmental Impact Assessment Review, 95, 106808.
08.01.2026
