A decentralized IoT and blockchain-based architecture for general-purpose vehicle speed data collection: a case study of Tanzania’s highways

Speed detection in Tanzania is predominantly conducted manually, particularly using logbooks and handheld speed guns. This approach is time-consuming for both road users and enforcement officers, demands considerable human efforts, and incurs significant operational costs for authorities, often compromising its effectiveness due to environmental factors such as weather conditions [1]. Furthermore, the manual methods are vulnerable to corruption, favoritism, and data forgery due to the absence of transparency and accountability enforcement mechanisms [1]. A notable example involves the dismissal of four police officers by the traffic police (TP) after it was discovered that they had deliberately deleted image evidence from police speed gun cameras, which were being used to document speeding violations for their gain [2]. In addition to manual enforcement, the Land Transport Regulatory Authority (LATRA) has implemented an automatic vehicle tracking system (VTS). This system is vehicle-based and limited in scope to intercity and regional public buses [3]. It involves the deployment of onboard Global Positioning System (GPS)-based speed monitoring devices that transmit data on speed violations and other metrics to vehicle owners and regulatory authorities [4, 5]. However, LATRA VTS faces integrity challenges; the devices are frequently tampered with to either disable them entirely or manipulate the data, including transmitting false information to authorities, as reported in Refs [6,7,8,9]. Furthermore, this system and others in the literature [10,11,12,13] rely on centralized architectures where a single central server handles processing, storage, and sharing of all data. This centralized setup inherently lacks transparency and is susceptible to data manipulation and tampering. Conversely, current vehicle speed detection methods in Tanzania mainly focus on enforcing speed limits, neglecting the broader opportunities and benefits of speed data across various sectors. Interviews with multiple organizations highlighted diverse needs for vehicle speed information. For example, the TP primarily use speed data for enforcement. At the same time, the National Road Safety Council (NRSC) depends on it to monitor TP activities and advise the government on road safety policies. Similarly, the road safety ambassadors (RSA) leverage speed data to increase public awareness about road safety. Bus and truck owners utilize it to manage responsible vehicle operation, and the Tanzania National Roads Agency (TANROADS) identifies key applications such as designing safer roads, planning maintenance schedules, and prioritizing repairs. Additionally, unions for bus and truck drivers rely on speed records to resolve disputes involving TP officers. Academic and research organizations such as the National Institute of Transport (NIT), Non-Governmental Organizations (NGOs) like the Tanzania Roads Association (TARA), insurance regulators like the Tanzania Insurance Regulatory Authority (TIRA), and anti-corruption agencies like the Prevention and Combating of Corruption Bureau (PCCB) could also derive significant benefits from access to reliable vehicle speed data. Despite this broad and increasing demand, there appears to be a lack of literature addressing vehicle speed data collection for general-purpose use. Existing research is narrowly focused, primarily on speed limit enforcement [14,15,16] or traffic management applications [17,18,19,20,21,22].

Internet of things (IoT) and blockchain technologies offer promising solutions to overcome these limitations of current vehicle speed detection systems. The IoT refers to a network of physical objects equipped with sensors, actuators, controllers, and communication interfaces, enabling the monitoring and control of environmental and physical conditions [23]. Usually, cloud computing (CC) has been used to support data storage and processing in IoT systems. However, the centralized cloud structure poses risks to security and trust, which raise concerns about data integrity, privacy, and trustworthiness [24]. Therefore, decentralized methods for managing IoT data are increasingly sought to improve transparency and security [25, 26]. Blockchain technology has emerged as a promising answer to many of these challenges. Using distributed ledger technology (DLT), blockchain provides data immutability, transparency, and trust through cryptographic processes [27,28,29]. As a result, the integration of IoT and blockchain technologies has garnered significant research attention [30]. Still, merging these technologies is complex, primarily due to IoT devices’ limited resources, such as energy, storage, and computing power, compared with the demanding resource requirements of blockchain systems in computation, storage, and energy use [26, 31, 32]. These disparities create critical challenges related to scalability, latency, energy efficiency, interoperability, and cost [33]. Additionally, no universally accepted architecture currently exists for integrating IoT and blockchain technologies [34]. Most existing research focuses on specific applications such as the Marine Internet of Things (MIoT) [34], healthcare [35,36,37], microgrids [38], accounting [39], supply chains [40], and cybersecurity [41]. However, no studies have specifically addressed the integration of IoT and blockchain technologies within vehicle speed detection. This gap highlights the need for further research, particularly in light of ongoing concerns on efficiency, scalability, transparency, and data security in existing systems. The main objective of this study is to develop a new architecture for integrating IoT and blockchain technologies to create an automated, secure, and transparent vehicle speed detection system suitable for Tanzania. The originality of this study is shown through three main contributions: (1) a framework for collecting and sharing general purpose vehicle speed data, enabling secure access for multiple stakeholders based on their privileges; (2) a decentralized architecture designed to improve transparency and reduce current risks related to centralization; and (3) the first known implementation of IoT and blockchain integration specifically for vehicle speed detection domain. Table 1 illustrates how the research questions are aligned with these contributions.

The rest of the paper is organized as follows: related works are presented in Section 2, the proposed methodology is presented in Section 3, the results are presented in Section 4, a discussion of the results is given in Section 5, and the conclusion of this study is given in Section 6.

Several IoT-blockchain integration architectures, ranging from generic frameworks to domain-specific solutions, have been proposed in the literature. Table 2 provides a comparative analysis of these architectures alongside the proposed model, evaluated based on key aspects such as architectural type, IoT device types, data processing and storage methods, application domains, blockchain parameters, and associated limitations. For example, Pavithran et al. [42] introduced a generic IoT and blockchain integration framework, highlighting essential factors for consideration, which laid the groundwork for many subsequent research efforts. However, despite this and similar efforts, there is currently no universally accepted architectural standard. Chacko et al. [33] presented a two-layer agricultural framework for multi-stakeholder interaction using blockchain across the agricultural supply chain. Yet, their framework encounters several deployment challenges: (i) full-node implementations are impractical in farm environments due to power constraints, limited connectivity, and security issues; (ii) the high cost of deploying numerous full-node devices makes the solution unsuitable for small-scale farmers; and (iii) the absence of an application layer leaves stakeholder interaction undefined. Aliyu and Liu [43] introduced a blockchain-based smart farm security framework with serverless services for processing event data and updating the Ethereum blockchain. While the three-layer architecture supports lightweight IoT devices, it also omits an application layer. A two-layer architecture for local energy trading is proposed by Condon et al. [44], but it relies on centralized storage, contradicts blockchain’s decentralization principles, and lacks a defined application interface for user interaction. A four-layer architecture utilizing hyperledger fabric (HLF) as a Blockchain-as-a-Service solution is presented by Eghmazi et al. [45] to address IoT security and privacy issues. Kumar and Sharma [46] introduced a three-layer model (IoT, off-chain, and blockchain), which integrates MongoDB for off-chain storage and HLF for ensuring data integrity in environmental monitoring applications. However, the application layer remains undefined. Deep et al. [24] introduced a two-layer architecture model, which enforces authentication and access control to enable secure environmental data sharing in a heterogeneous IoT setup, though its description lacks clarity, making interpretation of its structure and implementation difficult. The model introduced by Singal et al. [47] focuses on Ethereum-based IoT integration for secure supply chain traceability, but it does not include a comprehensive architectural design, instead providing only entity-relationship (ER) and flowchart diagrams. Similarly, Swati et al. [48] proposed a three-layer architecture for a blockchain-enabled IoT health monitoring system, utilizing both private and consortium blockchains for data confidentiality and collaboration among healthcare providers. A comparable study [35] introduced a three-layer architecture (IoT, data collector, and blockchain) for secure and privacy-preserving pharmaceutical data sharing in diabetes research, but it does not include a proof of concept (PoC) prototype. Furthermore, Dange and Nitnaware [49] suggested an IoT blockchain-based data-sharing model designed to guarantee the integrity of healthcare data exchanged among multiple stakeholders. Rather than storing all data on the blockchain, this model uses blockchain technology solely to verify data integrity. However, the approach appears complex and introduces significant overhead during the integrity verification process. These architectures cover multiple domains, including agriculture, energy, security, environment, supply chain, and healthcare. To our knowledge, no previous research specifically addresses the integration of IoT and blockchain technologies for vehicle speed monitoring systems. This study has effectively employed the HLF channels feature to create a consortium blockchain composed of organizations with different privileges. Moreover, a common limitation across most architectures is the lack of an application layer that defines user interaction. Additionally, evaluation efforts have mainly focused on performance metrics, with insufficient attention to other important aspects such as execution modes, functionality, and system integration.

The design decision factors outlined by Pavithran et al. [42] and Chacko et al. [50] provided guidelines for integrating IoT and blockchain technologies. These factors were adopted because they offered a structured framework for supporting the design decision-making process to create an effective and efficient architecture for integrating IoT and blockchain technologies. Each factor relevant to the merging of IoT and blockchain technologies within the systems building block was carefully analyzed and evaluated in the context of vehicle speed detection, considering the environmental factors, and both functional and non-functional requirements as presented in Table 3.

This study proposes a four-layer architecture design formed by the IoT, cloud, blockchain, and application layers, as illustrated in Figure 1. The IoT layer comprises lightweight sensor nodes. A typical node includes a radar sensor for measuring vehicle speed, a camera for capturing vehicle images, and a microcontroller (MCU) to process and transmit speed and image data to the cloud layer. To facilitate the transmission of data, the node utilizes a communication module such as a Wi-Fi module or a cellular modem. Additionally, an SD card may be incorporated to temporarily store data as a backup.

Figure 1:

Proposed architecture design for integrating IoT and blockchain technologies for vehicle speed detection systems. ANPR, automatic number plate recognition; IoT, Internet of things.

The cloud layer serves two primary functions: first, it handles data preprocessing tasks, which involve extracting number plate details from captured images using computer vision (CV) techniques, specifically automatic number plate recognition (ANPR), and categorizing vehicles based on their types through machine learning (ML) algorithms. Second, it facilitates the interconnection between the resource-constrained IoT nodes with the resource-intensive blockchain network. The blockchain layer is responsible for data storage across stakeholders. All data are securely recorded in a permissioned blockchain network, ensuring that only authorized organizations forming the consortium blockchain network have access. Additionally, because the proposed architecture is designed to be general-purpose, the HLF channel feature is used to separate organizations having different privileges on the data, while the application layer is responsible for handling application-specific data processing and sharing. Each organization within the blockchain network may deploy its application systems, such as web or mobile applications, tailored to its specific needs. As different blockchain channels serve distinct organizations, the data processing and visualization requirements are expected to vary according to the business objectives and operational demands of each entity. This approach ensures that each organization receives relevant, customized insights while maintaining secure, efficient, and transparent data sharing across the network.

This study proposes an architecture design to demonstrate how IoT-blockchain integration can be leveraged to implement a smart, secure, and transparent system for improving vehicle speed data collection in Tanzania. The first distinguishing feature of the proposed architecture is that, to the authors’ understanding, it is the first of its kind to exploit such technological integration in the vehicle speed data collection domain. Existing works focus on other domains such as health [35,36,37, 48, 49], supply chain [40, 47], agriculture [33, 43], accounting [39], and microgrids [38]. It features an innovative distribution of data collection, processing, storage, and sharing functionalities across the four layers of the architecture. The design effectively leverages the capabilities of IoT, artificial intelligence (AI), and blockchain, enabling the achievement of critical factors such as efficiency, transparency, and security. Furthermore, it demonstrates a streamlined yet effective combination of CC and blockchain technologies. CC is a well-established and mature technology to be replaced by blockchain. While it may be feasible to implement the solution entirely using either blockchain or CC, their combination can enhance the overall quality of the final product. The proposed architecture effectively harnesses the strengths of each integrated technology. CC serves as a vital intermediary between the IoT and blockchain layers, performing essential preprocessing tasks that would be too resource-intensive for IoT devices and too complex or costly to execute directly on the blockchain. This cloud layer offers several strategic advantages. It reduces system complexity by centralizing data preprocessing, thereby offloading computational demands from blockchain and IoT nodes. This enables the use of lightweight IoT nodes (typically comprising a radar sensor and an ESP32, which integrates the camera, MCU, Wi-Fi, and SD card into one module) while ensuring efficient integration with the resource-intensive blockchain infrastructure. Also, it significantly lowers costs by eliminating the need to equip every IoT or blockchain node with advanced capabilities such as ML and CV for performing ANPR and vehicle classification. Additionally, it enhances performance and optimizes resource utilization by avoiding redundant execution of tasks across nodes, as each task is performed only once within the cloud layer. These factors collectively contribute to the physical scalability of the system, allowing the number of IoT nodes to be easily increased as needed without affecting the existing physical design. Moreover, this architecture design may use the already existing cloud infrastructure provided by the Tanzania national e-government agency (e-GA), further minimizing operational costs. Second, the proposed design is particularly suited for deployment in rural and low-connectivity areas. The ESP32 module used in the IoT node supports seamless integration with cellular IoT modules such as long term evolution for machines (LTE-M) and NB-IoT through standard interfaces such as universal asynchronous receiver/transmitter (UART) and serial peripheral interface (SPI). Unlike works like [33, 43] which uses Wi-Fi, cellular IoT is a low power wide area network (LPWAN) technology built on Global System for Mobile Communication (GSM) infrastructure, the most widely deployed wireless communication system globally. It offers extended coverage and reliable connectivity, making it ideal for remote areas. Alternatively, Long Range Wide Area Network (LoRaWAN) may also be considered as a viable option, provided that the application’s data rate requirements can be satisfied.

The third distinguishing factor is the capability of collecting general-purpose vehicle speed data, compared with existing studies that primarily focus on speed limit law enforcement and traffic management. To resolve this, the design employs HLF channels to ensure data confidentiality and privacy for different categories of organizations. Different channels are implemented for various organizational groups, thereby enforcing fine-grained access control and authorization. This approach has enabled the design of a transparent and yet secure solution that upholds the privacy and security of data, addressing weaknesses of the existing centralized LATRA VTS [4, 5] and other systems [16, 50,51,52,53,54] which are prone to data manipulation and tampering. Table 6 presents the spoofing, tampering, repudiation, information disclosure, denial of service, and elevation of privilege (STRIDE) threat model illustrating the architecture’s resilience to attacks, and this model is structured into four layers. This model is adopted because it is the most mature modeling method to be used to assess the architecture layers against the main security properties [55].

Another distinctive feature of the proposed architecture lies in its application layer, which allows for application-specific data processing tailored to the diverse needs of different organizations. The system’s capability to collect vehicle speed data that serves multiple purposes across various sectors makes it highly adaptable and scalable. Each stakeholder can independently implement their interface, whether as a mobile or web application, depending on their unique business requirements and operational environments, as compared with references [45, 49], which are application layer specific, while works like [33, 43, 44, 46] did not include the application layer in their design. For example, in the context of law enforcement, police officers can utilize mobile applications for field operations. These may be used by traffic officers, patrol units, or other enforcement personnel to access real-time speed data and take immediate action. Conversely, web-based applications are more suited for office environments, where higher-ranking officers, such as those from the NRSC, may analyze aggregated data to generate reports, monitor trends, investigate incidents, and support decision-making. These applications can also be used to inform policy development, raise public awareness, and promote road safety research. From the transport sector perspective, bus and truck owners can implement mobile or web-based platforms to monitor driver behavior and ensure compliance with safety regulations. Similarly, the national road agency can analyze speed data for a range of critical functions, including the design of safer roads, conducting road safety audits, optimizing the placement of traffic signs and speed limits, and collaborating with other entities during accident investigations.

Another unique aspect of this study is its extensive evaluation scope, which went beyond conventional performance testing to include integration testing. The purpose of the integration testing was to verify the functionality of individual subsystems and ensure their smooth interaction for reliable end-to-end data delivery. Regarding performance evaluation, this study not only measured core metrics such as latency, error rate, and throughput as done in references [46, 49] but also examined the effectiveness of different execution strategies: sequential, parallel (multi-threaded), and asynchronous modes. These approaches were assessed to identify which of them delivered the best performance under various conditions. Specifically, performance-based tests evaluated end-to-end latency and error rates, offering insights into the system’s scalability, responsiveness, reliability, accuracy, robustness, and fault tolerance. Throughput tests analyzed data transmission efficiency across the system. All assessments were performed under two different configurations to explore how changes in the NR and the NN affect overall system performance. The results based on NR are shown in Figure 12A for sequential mode, Figure 12B for parallel mode, and Figure 12C for asynchronous mode. In sequential execution, latency ranged from 2,090 ms to 2,169 ms, indicating that increasing NR has little effect on latency. However, the overall latency remains high, likely due to the blocking nature of sequential execution, where each task waits for the previous one to finish. By contrast, parallel execution demonstrated greater efficiency and stability, with latency from 1,020 ms to 1,070 ms and an average of around 1,043 ms. This is nearly a 50% reduction compared with the averages for sequential (2,155 ms) and asynchronous (2,220 ms) modes. Minor variations in parallel latency probably stem from small scheduling overheads or payload size fluctuations, rather than inherent algorithm inefficiencies. Interestingly, the asynchronous mode, which is usually expected to outperform sequential execution because of its non-blocking nature, showed higher and more variable latency. This unexpected result might be due to factors such as network overhead, event loop contention, or managing callbacks with many concurrent tasks. While asynchronous execution is ideal for I/O-bound workloads and parallelism is better for central processing unit (CPU)-bound tasks, this project’s workload appears to be a balanced mix of both. For example, on the IoT (ESP32) side, operations such as generating speed data, encoding images in base64, and sending HTTP requests are mainly I/O-bound. On the cloud (Flask server), incoming requests undergo lightweight preprocessing (e.g., vehicle type and license plate detection), database management, and data transmission to the blockchain, mixing I/O and CPU-bound processes. On the blockchain, data are received and committed through chaincode, involving validation, network consensus, and storage more CPU-intensive tasks. Given this mixed workload, the parallel execution mode offers the best performance, with the lowest average latency and the highest throughput, as shown in Figure 13. Furthermore, throughput results (Figure 12) show that parallel mode achieves 0.96 records/s (Figure 12B), more than double that of sequential (0.46 records/s, Figure 12A) and asynchronous (0.45 records/s, Figure 12C) modes. Regarding error rates, all three modes perform acceptably: sequential: 0.02%, parallel: 0.07%, and asynchronous: 0.21%. These findings confirm that parallel execution provides the best balance between latency and throughput. Regarding how the NN affects performance, the trend observed in the parallel execution mode in Figure 15 shows a significant drop in latency after the initial data point, decreasing from 4,768 ms to around 2,500 ms. This early spike may be caused by delays in thread scheduling or resource initialization overhead. The subsequent measurements show a consistent reduction of about 50% in latency compared with the other execution modes. The average latencies recorded were 2,929.37 ms for the parallel mode, 4,792.86 ms for the asynchronous mode, and 4,786.65 ms for the sequential mode. This notable improvement in the parallel mode likely results from using two dedicated threads that run simultaneously, effectively simulating independent sequential or asynchronous processes. In terms of throughput, the parallel mode achieved the highest overall average at 0.36 records/s, as shown in Figure 14B, surpassing both the sequential and asynchronous modes, which each averaged 0.21 records/s, according to Figures 14A and 14B. Regarding the error rate, the parallel approach had the highest average at 7.48% in Figure 14B, followed by sequential at 4.92% in Figure 14A, and asynchronous at 2.24% in Figure 14C. However, it is worth noting that the error rate is mainly influenced by network quality rather than the execution method. Overall, the parallel execution mode demonstrated the most efficient performance, effectively balancing latency and throughput. Sequential execution was the slowest, with a final latency of 5,263 ms, which probably indicates limited scalability as workload increases. Interestingly, the asynchronous mode did not outperform the parallel mode, possibly due to I/O bottlenecks or the overhead linked to non-blocking operations. Overall, the evaluation results are very promising across all execution modes. Performance measurements in a real-world environment, particularly in terms of latency, error rate, and throughput, show the architecture’s ability to handle increasing numbers of nodes without significant performance loss. These findings highlight the architecture’s scalability, reliability, and readiness for practical deployment. Table 7 provides the performance, scalability, purpose, transparency, privacy, and security comparison against the existing centralized designs for vehicle speed monitoring.

In Tanzania, speed detection is still mainly done manually, a method that is inefficient, labor-intensive, costly, and highly affected by environmental factors such as weather. These limitations make manual systems vulnerable to corruption, favoritism, and data forgery. Existing automated vehicle-based systems are limited in scope, mainly targeting intercity and regional public buses. Furthermore, these systems are often tampered with to alter speed data. They are mostly restricted to enforcing speed limits, overlooking the broader potential of speed data for transportation planning, road safety, insurance, and policy development. Additionally, current solutions depend on centralized architectures, making them vulnerable to data manipulation and single points of failure. To address these issues, this study introduces a novel four-layer decentralized architecture that combines IoT and blockchain technologies to develop an automated, transparent, and secure vehicle speed detection system suited to the Tanzanian context. The proposed system shows promising performance and scalability across various evaluation metrics. Key contributions and findings include as follows:

• Demonstrating the feasibility of using IoT and blockchain technologies to create a transparent, secure, and automated vehicle speed data collection system.

• Pioneering the integration of IoT and blockchain in the vehicle speed detection field, representing, to our knowledge, the first such effort.

• Designing a versatile architecture for collecting vehicle speed data applicable across different domains.

• Utilizing HLF channels to ensure confidentiality and privacy in a multi-organizational blockchain environment.

• Developing a cost-effective and scalable design that uses lightweight IoT nodes suitable for rural and resource-limited settings.

Future works could explore leveraging the proposed architecture to implement transparent and secure vehicle speed data collection, as well as integrating 5G technology to enhance system performance.