Intelligent transportation systems (ITSs) leverage a network of interconnected infrastructures utilizing advanced technologies to improve traffic management and safety. Federated learning (FL) has emerged as a pivotal method within ITSs, enabling decentralized collaborative model training without direct data sharing, thus preserving privacy and enhancing system efficiency. This article explores the integration of FL in ITSs, focusing on FL’s application in traffic flow prediction, trajectory prediction, parking space estimation, and traffic target recognition. Despite its potential, FL deployment faces challenges, including data heterogeneity, communication and bandwidth constraints, and resource limitations on edge devices. Addressing these challenges is crucial for realizing the full potential of FL in ITSs. This article provides a comprehensive survey of existing FL implementations in ITSs, discusses inherent challenges, and outlines future research directions aimed at overcoming these obstacles.

This paper proposes load-balancing algorithm on the basis of the Extended Dijkstra’s shortest path algorithm for Software Defined Networking (SDN). The Extended Dijkstra’s algorithm considers not only the edge weights, but also the node weights to find the nearest server for a requesting client. The proposed algorithm also considers the link load in order to avoid congestion. We use Pyretic to implement the proposed algorithm and compare it with related ones under the Abilene network topology with the Mininet emulation tool. As shown by the comparisons, the proposed algorithm outperforms the others in term of the network end-to-end latency, throughput and response time at the expense of a little heavier computation load and more memory usage on the SDN controller.

In the context of edge computing, a cluster of small single-board computers like Raspberry Pi could serve as robust servers. These clusters not only offer robust server capabilities but also exhibit a remarkable versatility in handling incoming data streams from a multitude of sensors within the Internet of Things (IoT) ecosystem, particularly in underserved rural areas. Simultaneously, they can seamlessly double up as servers for web-based applications tailored to the specific requirements of small businesses. However, the operational context of such an edge computing cluster can present challenges. For instance, dynamic load fluctuations, ranging from high to low demands, may lead to performance service degradation or underutilized services. This is a typical problem in distributed computing environments, where the heterogeneity of devices, dynamic conditions, and reliability of connections can create scalability issues. This paper addresses these challenges through a selective set of integrated software suites aiming to autoscale an edge computing cluster. The software suites consist of a lightweight Kubernetes distribution called K3s, with an automation framework executed through Ansible. Rigorous testing, primarily focused on web-based applications, has showcased the efficacy of this approach. A compelling comparison has been drawn between this optimized edge computing setup and conventional desktop-based servers, emphasizing superior power efficiency and commendable performance levels. The service scaling can reduce power consumption by up to 45%.

The Internet of Things (IoT) brings millions of devices, in the form of sensor, actuator, or embedded electronic devices interconnected through the Internet. Monitoring abundant of IoT devices is a big challenge, especially in a constrained condition of sensor nodes (e.g. limited CPU, memory, and battery). Therefore, a good monitoring system with an efficient monitoring protocol is often needed to compensate the limited resources of sensor nodes. SNMP is a protocol that widely used for a monitoring system. However, SNMP was developed without taking into account of devices with limited resources. This paper presents a proposed prototype of a light weight monitoring system based on agent-manager model. The testing result shows that the developed prototype system generates smaller packet sizes than the SNMP-based monitoring system on the same function and shows a low CPU and memory usage in monitoring thousands of devices.

Internet of things (IoT) is a complex system with few best practices in building ones, especially on handling real-time communication between IoT devices to the Internet. A framework is often used to fasten building IoT system. This paper present Mobile Internet of Things (MIOT), a framework which use a smartphone as the main component to handle communication between IoT device and the internet. A smartphone is used as the communication gateway (relay) for IoT devices and not as the IoT controller as in common Smartphone-IoT approach. For evaluation purpose, two implementations of IoT prototype scenario is built, an environmental monitoring and a remote controller (RC) car. The experiment shows a quick and easy deployment of IoT system. The Environment Monitoring able to send data to the server in real-time, and control The RC Car with a reasonable response time. The experiment on 200 ms interval between each packet, shows that MIOT Framework has round-trip latency between MIOT Server and IoT hardware for ≈ 88.007 ms. The addition of smartphone as the main component in the framework (MIOT Apps) contribute to additional latency ≈ 13.145 ms. Using a Smartphone as a gateway for IoT in MIOT Framework is possible and promising. It can be used as a best practice to develop a reliable IoT system which reduces time, effort, and learning overhead on building IoT systems.

Predicting arrival times of buses is a key challenge in the context of building intelligent public transportation systems. The bus arrival time is the primary information for providing passengers with an accurate information system that can reduce passenger waiting times. In this paper, we used the normal distribution method to the random of travel times data in a bus line number 243 in Taipei area. In developing the models, data were collected from Taipei Bus Company. A normal distribution method used for predicting the bus arrival time in bus stop to ensure users not to miss the bus, and compare the result with the existing application. The result of our experiment showed that our proposed method has a better prediction than existing application, with the probability user not to miss the bus in peak time is 93% and in normal time is 85%, greater than from the existing application with the 65% probability in peak time, and 70% in normal time.