In smart cities, infrastructure-side sensors are already used to increase safety, to mitigate traffic congestion and to reduce pollution caused by vehicles. In the future, infrastructural safeguarding is expected to get a large potential due to availability of advanced infrastructure sensors (camera,radar,lidar) and vehicle-to-infrastructure (V2I) communications. In principle, installations of infrastructure-side sensor systems can be divided into two main categories as temporary and permanent. Permanent systems are typically installed for highly used part of urban environments such as city centers, whereas temporary systems are setup for events in cities (concerts,sports games,etc.) with a temporally high density of traffic participants. For both categories different networking options (wired,wireless) as well as sensor data fusion methods (centralized, decentralized data fusion) are possible in order to meet requirements of the specific use case. In this paper, we consider two main architecture options: (1) wired networking with centralized data fusion (centralize architecture) and (2) wireless networking with decentralized data fusion (decentralized architecture).
In the centralized network architecture, the we assume that raw data from multiple sensors will be transmitted to a central processor unit (master). In this architecture, a single node handles the data processing and the fusion process. In the decentralized network architecture, every sensor is equipped with a computing and a communication component. Sensors use a co-located processor unit (slave) to, preprocess data and to transmit these data to a master node. In both architecture options, the master node will be responsible for combining all inputs from multiple sensors to form a common estimate for future state. A centralized architecture requires high-bandwidth connectivity between the sensors and the master, whereas the preprocessing in the decentralized approach has lower demands on the networking bandwidth.
Considering the two architectures the synchronization of the sensor time strongly impacts the state estimation. Sensor nodes need to synchronize their operation and collaborate to accomplish the sensing task. For example, in order to track a vehicle, sensors need to report the location and detection time of vehicle to master node. Then, the master node combines the information to estimate the location and velocity of the vehicle. Evidently, if the sensors do not have a common timescale, the state estimation will be inaccurate. Other aspect is how old the fused data (estimation) are before it delivered to road participants. When road participants receive a current estimation at the timestamp T2, these data represent state at time T1. The time difference between T1 and T2 depends on primarily on the network delay and process time of sensor data. For performance evaluation of the network options with respect to time synchronization the data processing will be separated as an additional source of delay.
In this paper, we study the two architecture options for a smart road infrastructure. The study is based on prototype system using the Robot Operating System (ROS) and cameras, whereas also other sensors (radar,LiDAR) are considered. We specifically assess the impact of sensor time synchronization on the reliability of the sensor data fusion and evaluate the latency between the acquisition of sensor data and reception of the data by the road participants.
Mr. Numan Senel, Technische Hochschule Ingolstadt, GERMANY; Prof. Dr. Gordon Elger, Technische Hochschule Ingolstadt, GERMANY; Prof. Dr. Andreas Festag, Fraunhofer IVI Applied Research Center for Connected Mobility and Infrastructure, GERMANY