A Graph-Based Methodology for the Sensorless Estimation of Road Traffic Profiles

Traffic forecasting models rely on data that needs to be sensed, processed, and stored. This requires the deployment and maintenance of traffic sensing infrastructure, often leading to unaffordable monetary costs. The lack of sensed locations can be complemented with synthetic data simulations that further lower the economical investment needed for traffic monitoring. One of the most common data generative approaches consists of producing real-like traffic patterns, according to data distributions from analogous roads. The process of detecting roads with similar traffic is the key point of these systems. However, without collecting data at the target location no flow metrics can be employed for this similarity-based search. We present a method to discover locations among those with available traffic data by inspecting topological features. These features are extracted from domain-specific knowledge as numerical representations (embeddings) to compare different locations and eventually find roads with analogous daily traffic profiles based on the similarity between embeddings. The performance of this novel selection system is examined and compared to simpler traffic estimation approaches. After finding a similar source of data, a generative method is used to synthesize traffic profiles. Depending on the resemblance of the traffic behavior at the sensed road, the generation method can be fed with data from one road only. Several generation approaches are analyzed in terms of the precision of the synthesized samples. Above all, this work intends to stimulate further research efforts towards enhancing the quality of synthetic traffic samples and thereby, reducing the need for sensing infrastructure.

I. Introduction

One of the main goals of traffic management is to provide a safe traffic infrastructure, where most of the potential traffic congestion scenarios and car collisions are avoided. This issue can be addressed by modeling the traffic behavior of the road network and by developing predictive methods that allow forecasting metrics that characterize traffic, such as flow, speed, or travel time [1]. This process enables the implementation of operational measures for decision making (e.g., redirecting traffic flow to other alternative routes). These traffic forecasting models are usually built from past traffic observations. However, in practice traffic data acquisition systems cannot be deployed over every link of a road, mainly due to the high costs of deployment and maintenance of such sensing equipment. On many occasions this issue has been addressed by deploying temporal sensors that provide measurements for certain locations of interest during a limited period of time. Nevertheless, a proper characterization of the traffic behavior under a variety of circumstances (e.g., events or holidays) requires real traffic measurements over more dilated periods [2]. Thus, real traffic data are not available for every road link, nor are they collected for the time needed for a thorough characterization. Indeed, in some cases placing a sensor is not feasible due to manifold reasons. If possible, another concern arises from the long latency needed to collect sufficient data for the subsequent model training. Since the performance of traffic forecasting models is constrained to both data quality and quantity, data availability is arguably the practical bottleneck for a successful traffic network characterization.