Abstract
In some places, the continuously increasing road traffic will soon exhaust the capacity of existing traffic infrastructure unless appropriate measures are taken. Especially in urban environments with a high density of residential and commercial properties, the infrastructure is highly utilized or overloaded during peak hours. Since structural measures are often not possible or only at great expense, a practical solution to counter this issue is to optimize the infrastructure utilization and the control of traffic flows. For this purpose, the widely installed Internet of Things (IoT)-powered Intelligent Traffic Systems (ITS) can be used,which enable automated detection and high-precision classification of different road users and thus transform the infrastructure into a datadriven Cyber-Physical System (CPS). Although various sensor systems have been proposed, they fulfill only subsets of the requirements, including accuracy, cost-efficiency, privacy preservation, and robustness. One approach that meets those requirements is a novel radio-based sensor system, of which we present two variants in this contribution. The system’s fundamental idea is to exploit radio-based fingerprints of road users-multi-dimensional and characteristic attenuation patterns of several radio links-for detection and classification. One of the presented system variants additionally evaluates high-precision channel information extracted fromWireless LAN (WLAN) Channel State Information (CSI) or Ultra-Wideband (UWB) Channel Impulse Response (CIR) data. The proposed solution benefits from increased robustness against a wide range of interferences, e. g., poor visibility due to bad weather conditions. Moreover, the system exclusively uses embedded microcontroller units (MCUs) and radio technologies, allowing compact and cost-efficient installations in rural and dense downtown areas. We have performed comprehensive field measurement campaigns and machine learning-enabled analyses that confirm the presented approach’s high suitability for different requirements and application scenarios. In this regard, we have evaluated multiple applications, including the comparatively simple detection of road users and the fine-grained classifications of several vehicle classes. For instance, the proposed systems achieve more than 99%for binary classification and 93.83%for differentiating seven vehicle types.
