Pub Date : 2024-08-26DOI: 10.1109/OJITS.2024.3449698
Xingyu Liu;Yuanfeng Chu;Yiheng Hu;Nan Zhao
Road target detection is essential for enhancing vehicle safety, increasing operational efficiency, and optimizing user experience. It also forms a crucial part of autonomous driving and intelligent monitoring systems. However, current technologies face significant limitations in multi-level feature fusion and the accurate identification of key targets in complex data environments. To address these challenges, this paper proposes an innovative algorithmic model called BiFPN GAM SimC2f-YOLO (BGS-YOLO), aimed at improving detection performance. Initially, this paper employs the Bidirectional Feature Pyramid Network (BiFPN) to effectively integrate multi-level features. This integration overcomes the limitations in feature extraction and recognition found in existing target detection algorithms. Following this, this paper introduces the Global Attention Module (GAM), which markedly improves the efficiency and accuracy of extracting key target information in complex data environments. Additionally, this paper innovatively designs the SimAM-C2f (SimC2f) network, further advancing feature expressiveness and fusion efficiency. Experiments on the public COCO dataset demonstrate that the BGS-YOLO model significantly outperforms the existing YOLOv8n model. Notably, it shows a 7.3% increase in mean average precision (mAP) and a 2.4% improvement in accuracy. These results highlight the model’s high precision and swift response in detecting road targets in complex traffic scenarios. Consequently, the BGS-YOLO model has the potential to significantly enhance road safety and contribute to a considerable reduction in traffic accident rates.
{"title":"Enhancing Intelligent Road Target Monitoring: A Novel BGS-YOLO Approach Based on the YOLOv8 Algorithm","authors":"Xingyu Liu;Yuanfeng Chu;Yiheng Hu;Nan Zhao","doi":"10.1109/OJITS.2024.3449698","DOIUrl":"https://doi.org/10.1109/OJITS.2024.3449698","url":null,"abstract":"Road target detection is essential for enhancing vehicle safety, increasing operational efficiency, and optimizing user experience. It also forms a crucial part of autonomous driving and intelligent monitoring systems. However, current technologies face significant limitations in multi-level feature fusion and the accurate identification of key targets in complex data environments. To address these challenges, this paper proposes an innovative algorithmic model called BiFPN GAM SimC2f-YOLO (BGS-YOLO), aimed at improving detection performance. Initially, this paper employs the Bidirectional Feature Pyramid Network (BiFPN) to effectively integrate multi-level features. This integration overcomes the limitations in feature extraction and recognition found in existing target detection algorithms. Following this, this paper introduces the Global Attention Module (GAM), which markedly improves the efficiency and accuracy of extracting key target information in complex data environments. Additionally, this paper innovatively designs the SimAM-C2f (SimC2f) network, further advancing feature expressiveness and fusion efficiency. Experiments on the public COCO dataset demonstrate that the BGS-YOLO model significantly outperforms the existing YOLOv8n model. Notably, it shows a 7.3% increase in mean average precision (mAP) and a 2.4% improvement in accuracy. These results highlight the model’s high precision and swift response in detecting road targets in complex traffic scenarios. Consequently, the BGS-YOLO model has the potential to significantly enhance road safety and contribute to a considerable reduction in traffic accident rates.","PeriodicalId":100631,"journal":{"name":"IEEE Open Journal of Intelligent Transportation Systems","volume":"5 ","pages":"509-519"},"PeriodicalIF":4.6,"publicationDate":"2024-08-26","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"https://ieeexplore.ieee.org/stamp/stamp.jsp?tp=&arnumber=10646366","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"142143833","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"OA","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Pub Date : 2024-08-15DOI: 10.1109/OJITS.2024.3444487
José Amendola;Linga Reddy Cenkeramaddi;Ajit Jha
Unmanned aerial vehicles, and special multirotor drones, have shown great relevance in a plethora of missions that require high affordance, field of view, and precision. Their limited payload capacity and autonomy make its landing a crucial task. Despite many attempts in the literature to address drone landing, challenges and open gaps still exist. Reinforcement Learning has gained notoriety in a variety of control problems, with recent proposals for drone landing applications. This work aims to present a systematic literature review on works employing Deep Reinforcement Learning for multirotor drone landing in both static and dynamic platforms. It also revisits Reinforcement Learning Algorithms, the main frameworks and simulators adopted for specific landing operations. The comprehensive analysis performed on reviewed works revealed that there are important untackled challenges when it comes to wind disturbances, unpredictability of moving landing targets, sensor latency, and sim-to-real gap. Finally, we present our critical analysis of how recent state-of-the-art deep learning concepts can be combined with reinforcement learning to leverage the latter in addressing the open gaps in future works.
{"title":"Drone Landing and Reinforcement Learning: State-of-Art, Challenges and Opportunities","authors":"José Amendola;Linga Reddy Cenkeramaddi;Ajit Jha","doi":"10.1109/OJITS.2024.3444487","DOIUrl":"https://doi.org/10.1109/OJITS.2024.3444487","url":null,"abstract":"Unmanned aerial vehicles, and special multirotor drones, have shown great relevance in a plethora of missions that require high affordance, field of view, and precision. Their limited payload capacity and autonomy make its landing a crucial task. Despite many attempts in the literature to address drone landing, challenges and open gaps still exist. Reinforcement Learning has gained notoriety in a variety of control problems, with recent proposals for drone landing applications. This work aims to present a systematic literature review on works employing Deep Reinforcement Learning for multirotor drone landing in both static and dynamic platforms. It also revisits Reinforcement Learning Algorithms, the main frameworks and simulators adopted for specific landing operations. The comprehensive analysis performed on reviewed works revealed that there are important untackled challenges when it comes to wind disturbances, unpredictability of moving landing targets, sensor latency, and sim-to-real gap. Finally, we present our critical analysis of how recent state-of-the-art deep learning concepts can be combined with reinforcement learning to leverage the latter in addressing the open gaps in future works.","PeriodicalId":100631,"journal":{"name":"IEEE Open Journal of Intelligent Transportation Systems","volume":"5 ","pages":"520-539"},"PeriodicalIF":4.6,"publicationDate":"2024-08-15","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"https://ieeexplore.ieee.org/stamp/stamp.jsp?tp=&arnumber=10637701","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"142174012","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"OA","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
This work proposes and investigates a solution approach to the urban air transportation network optimization problem, considering the perspectives of different stakeholders, including societal interests. Given logistic hub positions and a set of optimized paths connecting them pairwise, we aim for a Pareto-optimal and three-dimensional air corridor network structure. This work demonstrates a way to merge the given paths into a network and provides a framework to optimize the network further regarding multiple objectives. It proposes three objective functions that evaluate the network from the economic perspectives of network providers and users and the city residents’ social point of view. Using geospatial data from Frankfurt, Germany, we conducted different experiments including and excluding the social objective function under a varying input set of pre-optimized paths. Our analysis showed that taking social aspects into account results in traffic networks whose increase in social acceptance far outweighs the extra monetary costs. We conclude that it is beneficial to integrate social criteria into optimization problems when the solutions obtained are the basis for decisions in the area of conflict between the economy and human welfare.
{"title":"Multi-Objective Optimization of Urban Air Transportation Networks Under Social Considerations","authors":"Nikolas Hohmann;Sebastian Brulin;Jürgen Adamy;Markus Olhofer","doi":"10.1109/OJITS.2024.3443170","DOIUrl":"10.1109/OJITS.2024.3443170","url":null,"abstract":"This work proposes and investigates a solution approach to the urban air transportation network optimization problem, considering the perspectives of different stakeholders, including societal interests. Given logistic hub positions and a set of optimized paths connecting them pairwise, we aim for a Pareto-optimal and three-dimensional air corridor network structure. This work demonstrates a way to merge the given paths into a network and provides a framework to optimize the network further regarding multiple objectives. It proposes three objective functions that evaluate the network from the economic perspectives of network providers and users and the city residents’ social point of view. Using geospatial data from Frankfurt, Germany, we conducted different experiments including and excluding the social objective function under a varying input set of pre-optimized paths. Our analysis showed that taking social aspects into account results in traffic networks whose increase in social acceptance far outweighs the extra monetary costs. We conclude that it is beneficial to integrate social criteria into optimization problems when the solutions obtained are the basis for decisions in the area of conflict between the economy and human welfare.","PeriodicalId":100631,"journal":{"name":"IEEE Open Journal of Intelligent Transportation Systems","volume":"5 ","pages":"589-602"},"PeriodicalIF":4.6,"publicationDate":"2024-08-13","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"https://ieeexplore.ieee.org/stamp/stamp.jsp?tp=&arnumber=10634851","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"142184249","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"OA","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Pub Date : 2024-08-09DOI: 10.1109/OJITS.2024.3441048
Franca Rocco Di Torrepadula;Sergio Di Martino;Nicola Mazzocca;Paolo Sannino
Smart cities include complex ICT ecosystems, whose definition requires the cooperation of several software systems. Among them, Intelligent Public Transportation Systems (IPTS) aim to effectively exploit public transit resources. Still, adopting an IPTS is non-trivial. Off-the-shelf IPTS are often tied to specific technologies and, thus, not easy to integrate within existing software ecosystems. Moreover, despite IPTS introduce several peculiar issues, there is a lack of domain-specific reference architectures, which would significantly ease the work of practitioners. To fill this gap, starting from the experience gained with the Hitachi Rail company in deploying a large-scale IPTS, we identify a set of requirements for IPTS, and propose a domain-specific reference architecture, compliant with these requirements, whose primary objective is facilitating and standardizing the design of IPTS, by providing guidelines to IPTS designers. Consequently, it eases also the interoperability among different IPTSs. As an example of an IPTS obtainable from the architecture, we present a solution currently deployed by Hitachi in a major Italian city. Still, being independent from the specific considered urban scenario, the architecture can be easily instantiated in different cities with similar needs. Finally, we discuss some research challenges which should be further investigated in this domain.
{"title":"A Reference Architecture for Data-Driven Intelligent Public Transportation Systems","authors":"Franca Rocco Di Torrepadula;Sergio Di Martino;Nicola Mazzocca;Paolo Sannino","doi":"10.1109/OJITS.2024.3441048","DOIUrl":"10.1109/OJITS.2024.3441048","url":null,"abstract":"Smart cities include complex ICT ecosystems, whose definition requires the cooperation of several software systems. Among them, Intelligent Public Transportation Systems (IPTS) aim to effectively exploit public transit resources. Still, adopting an IPTS is non-trivial. Off-the-shelf IPTS are often tied to specific technologies and, thus, not easy to integrate within existing software ecosystems. Moreover, despite IPTS introduce several peculiar issues, there is a lack of domain-specific reference architectures, which would significantly ease the work of practitioners. To fill this gap, starting from the experience gained with the Hitachi Rail company in deploying a large-scale IPTS, we identify a set of requirements for IPTS, and propose a domain-specific reference architecture, compliant with these requirements, whose primary objective is facilitating and standardizing the design of IPTS, by providing guidelines to IPTS designers. Consequently, it eases also the interoperability among different IPTSs. As an example of an IPTS obtainable from the architecture, we present a solution currently deployed by Hitachi in a major Italian city. Still, being independent from the specific considered urban scenario, the architecture can be easily instantiated in different cities with similar needs. Finally, we discuss some research challenges which should be further investigated in this domain.","PeriodicalId":100631,"journal":{"name":"IEEE Open Journal of Intelligent Transportation Systems","volume":"5 ","pages":"469-482"},"PeriodicalIF":4.6,"publicationDate":"2024-08-09","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"https://ieeexplore.ieee.org/stamp/stamp.jsp?tp=&arnumber=10632072","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"141944413","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"OA","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Pub Date : 2024-08-08DOI: 10.1109/OJITS.2024.3440876
Jueal Mia;M. Hadi Amini
Federated Learning is a decentralized machine learning technique that creates a global model by aggregating local models from multiple edge devices without a need to access the local data. However, due to the distributed nature of federated learning, there is a larger attack surface, making cyber-attack detection and defense challenging. Although prior works developed various defense strategies to address security issues in federated learning settings, most approaches fail to mitigate cyber-attacks due to the diverse characteristics of the attack, edge devices, and data distribution. To address this issue, this paper develops a hybrid privacy-preserving algorithm to safeguard federated learning methods against malicious attacks in Intelligent Transportation Systems, considering object detection as a downstream machine learning task. This algorithm involves the edge devices (e.g., autonomous vehicles) and road side units to collaboratively train their model while maintaining the privacy of their respective data. Furthermore, this hybrid algorithm provides robust security against data poisoning-based model replacement and inference attacks throughout the training phase. We evaluated our model using the CIFAR10 and LISA traffic light dataset, demonstrating its ability to mitigate malicious attacks with minimal impact on the performance of main tasks.
联盟学习是一种去中心化的机器学习技术，它通过聚合多个边缘设备的本地模型来创建全局模型，而无需访问本地数据。然而，由于联合学习的分布式特性，攻击面较大，使得网络攻击检测和防御具有挑战性。虽然之前的研究开发了各种防御策略来解决联合学习环境中的安全问题，但由于攻击、边缘设备和数据分布的不同特点，大多数方法都无法缓解网络攻击。为解决这一问题，本文开发了一种混合隐私保护算法，以保护联合学习方法免受智能交通系统中的恶意攻击，并将目标检测视为下游机器学习任务。该算法涉及边缘设备（如自动驾驶汽车）和路侧设备，在维护各自数据隐私的同时，协同训练其模型。此外，这种混合算法还能在整个训练阶段提供强大的安全性，防止基于数据中毒的模型替换和推理攻击。我们使用 CIFAR10 和 LISA 交通灯数据集对我们的模型进行了评估，证明它有能力在对主要任务性能影响最小的情况下缓解恶意攻击。
{"title":"A Secure Object Detection Technique for Intelligent Transportation Systems","authors":"Jueal Mia;M. Hadi Amini","doi":"10.1109/OJITS.2024.3440876","DOIUrl":"10.1109/OJITS.2024.3440876","url":null,"abstract":"Federated Learning is a decentralized machine learning technique that creates a global model by aggregating local models from multiple edge devices without a need to access the local data. However, due to the distributed nature of federated learning, there is a larger attack surface, making cyber-attack detection and defense challenging. Although prior works developed various defense strategies to address security issues in federated learning settings, most approaches fail to mitigate cyber-attacks due to the diverse characteristics of the attack, edge devices, and data distribution. To address this issue, this paper develops a hybrid privacy-preserving algorithm to safeguard federated learning methods against malicious attacks in Intelligent Transportation Systems, considering object detection as a downstream machine learning task. This algorithm involves the edge devices (e.g., autonomous vehicles) and road side units to collaboratively train their model while maintaining the privacy of their respective data. Furthermore, this hybrid algorithm provides robust security against data poisoning-based model replacement and inference attacks throughout the training phase. We evaluated our model using the CIFAR10 and LISA traffic light dataset, demonstrating its ability to mitigate malicious attacks with minimal impact on the performance of main tasks.","PeriodicalId":100631,"journal":{"name":"IEEE Open Journal of Intelligent Transportation Systems","volume":"5 ","pages":"495-508"},"PeriodicalIF":4.6,"publicationDate":"2024-08-08","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"https://ieeexplore.ieee.org/stamp/stamp.jsp?tp=&arnumber=10630660","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"141944414","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"OA","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
As the automotive industry evolves, integrating intelligent technologies and cooperative services in vehicular networks has become crucial to enhance road safety and autonomous driving capabilities. However, this integration can strain networks, particularly when exchanging a high volume of object information. This work studies the impact of the Collective Perception Messages (CPMs) size on the vehicular network performance. We introduce an algorithm aimed at optimizing the efficiency of extra object data inclusion in CPMs. The focus is on evaluating the vehicular network efficiency by selectively including extra objects within the available message space, strategically enhancing the transmission of more objects. This optimization not only reduces the need for constant CPM generation, but also maximizes the efficiency of information exchange. Using real-world vehicular data, this approach’s effectiveness in improving the Collective Perception Service (CPS) is demonstrated, showing a significant improvement when compared to traditional CPS standard: the proposed algorithm is capable of transmitting 14% more object information while using 2.6% fewer bytes. In addition, if we were to maintain the same number of bytes used in transmission as the CPS standard, our algorithm would result in a 23% increase in transmitted object information. Furthermore, the additional delay incurred by the algorithm is minimal, with an average of just 3 ms.
{"title":"Enhancing Vehicular Network Efficiency: The Impact of Object Data Inclusion in the Collective Perception Service","authors":"Andreia Figueiredo;Pedro Rito;Miguel Luís;Susana Sargento","doi":"10.1109/OJITS.2024.3437206","DOIUrl":"10.1109/OJITS.2024.3437206","url":null,"abstract":"As the automotive industry evolves, integrating intelligent technologies and cooperative services in vehicular networks has become crucial to enhance road safety and autonomous driving capabilities. However, this integration can strain networks, particularly when exchanging a high volume of object information. This work studies the impact of the Collective Perception Messages (CPMs) size on the vehicular network performance. We introduce an algorithm aimed at optimizing the efficiency of extra object data inclusion in CPMs. The focus is on evaluating the vehicular network efficiency by selectively including extra objects within the available message space, strategically enhancing the transmission of more objects. This optimization not only reduces the need for constant CPM generation, but also maximizes the efficiency of information exchange. Using real-world vehicular data, this approach’s effectiveness in improving the Collective Perception Service (CPS) is demonstrated, showing a significant improvement when compared to traditional CPS standard: the proposed algorithm is capable of transmitting 14% more object information while using 2.6% fewer bytes. In addition, if we were to maintain the same number of bytes used in transmission as the CPS standard, our algorithm would result in a 23% increase in transmitted object information. Furthermore, the additional delay incurred by the algorithm is minimal, with an average of just 3 ms.","PeriodicalId":100631,"journal":{"name":"IEEE Open Journal of Intelligent Transportation Systems","volume":"5 ","pages":"454-468"},"PeriodicalIF":4.6,"publicationDate":"2024-08-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"https://ieeexplore.ieee.org/stamp/stamp.jsp?tp=&arnumber=10620279","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"141880577","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"OA","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
The last few decades have witnessed a dramatic evolution of Artificial Intelligence (AI) algorithms, represented by Deep Neural Networks (DNNs), resulting in AI-enabled systems being significantly dominant in various fields, including robotics, healthcare, and mobility. AI-enabled systems are currently used even for safety-critical applications, including automated driving, where they encounter reliability challenges from both hardware (HW) and software (SW) perspectives. However, there is no effective technique available that can diagnose HW and SW of AI-enabled systems in real-time during operation. Therefore, this paper proposes an intelligent real-time diagnostic technique for detecting HW and SW anomalies of AI-enabled systems and continuously improving the SW quality during operation. The proposed technique can detect HW anomalies to avoid unexpected changes in AI parameters and subsequent AI performance degradation using single context data with a detection accuracy of more than 92%. The proposed technique can also detect SW anomalies and identify edge cases in real-time, which could result in performance degradation by more than 50% compared to normal conditions. The identified edge cases can be used to continuously enhance the SW quality. Experimental results show the effectiveness of the technique for practical applications and thus can contribute to realize reliable and improved AI-enabled systems.
{"title":"Real-Time Diagnostic Technique for AI-Enabled System","authors":"Hiroaki Itsuji;Takumi Uezono;Tadanobu Toba;Subrata Kumar Kundu","doi":"10.1109/OJITS.2024.3435712","DOIUrl":"10.1109/OJITS.2024.3435712","url":null,"abstract":"The last few decades have witnessed a dramatic evolution of Artificial Intelligence (AI) algorithms, represented by Deep Neural Networks (DNNs), resulting in AI-enabled systems being significantly dominant in various fields, including robotics, healthcare, and mobility. AI-enabled systems are currently used even for safety-critical applications, including automated driving, where they encounter reliability challenges from both hardware (HW) and software (SW) perspectives. However, there is no effective technique available that can diagnose HW and SW of AI-enabled systems in real-time during operation. Therefore, this paper proposes an intelligent real-time diagnostic technique for detecting HW and SW anomalies of AI-enabled systems and continuously improving the SW quality during operation. The proposed technique can detect HW anomalies to avoid unexpected changes in AI parameters and subsequent AI performance degradation using single context data with a detection accuracy of more than 92%. The proposed technique can also detect SW anomalies and identify edge cases in real-time, which could result in performance degradation by more than 50% compared to normal conditions. The identified edge cases can be used to continuously enhance the SW quality. Experimental results show the effectiveness of the technique for practical applications and thus can contribute to realize reliable and improved AI-enabled systems.","PeriodicalId":100631,"journal":{"name":"IEEE Open Journal of Intelligent Transportation Systems","volume":"5 ","pages":"483-494"},"PeriodicalIF":4.6,"publicationDate":"2024-07-30","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"https://ieeexplore.ieee.org/stamp/stamp.jsp?tp=&arnumber=10614714","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"141872302","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"OA","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Pub Date : 2024-07-29DOI: 10.1109/OJITS.2024.3435078
N. Neis;J. Beyerer
The lateral movement of vehicles within their lane under homogeneous traffic conditions is decisive for the range of vision of vehicle sensors. It significantly contributes to the maximum situational awareness an automated driving function can achieve. Given the integral role that simulations play in the validation of automated driving functions, the development of models that accurately describe the lateral movement of vehicles within their lane becomes essential. A few models have already been proposed in literature that address this task. Existing models, however, exhibit limitations when it comes to their usage for the virtual validation of automated driving functions such as the replication of general instead of driver-specific behavior and complex calibration methods. Furthermore, the metrics used for evaluation focus on measuring the accordance of the overall lateral offset and speed distribution and do not take into account the temporal course of the lateral offset profiles. Within this work, we introduce a two-level stochastic model addressing the identified limitations. We further present a strategy suitable for evaluating the low-level characteristics of the generated lateral offset profiles relevant for validating an automated driving function such as a cut-in detection function within simulations. The model’s capabilities are demonstrated based on five single driver datasets. It is shown that the model allows for efficient calibration, is able to replicate the behavior of these drivers, and is characterized by short runtimes. This makes it suitable for the virtual validation of automated driving functions.
{"title":"Efficiently Modeling Lateral Vehicle Movement Including its Temporal Interrelations Using a Two-Level Stochastic Model","authors":"N. Neis;J. Beyerer","doi":"10.1109/OJITS.2024.3435078","DOIUrl":"10.1109/OJITS.2024.3435078","url":null,"abstract":"The lateral movement of vehicles within their lane under homogeneous traffic conditions is decisive for the range of vision of vehicle sensors. It significantly contributes to the maximum situational awareness an automated driving function can achieve. Given the integral role that simulations play in the validation of automated driving functions, the development of models that accurately describe the lateral movement of vehicles within their lane becomes essential. A few models have already been proposed in literature that address this task. Existing models, however, exhibit limitations when it comes to their usage for the virtual validation of automated driving functions such as the replication of general instead of driver-specific behavior and complex calibration methods. Furthermore, the metrics used for evaluation focus on measuring the accordance of the overall lateral offset and speed distribution and do not take into account the temporal course of the lateral offset profiles. Within this work, we introduce a two-level stochastic model addressing the identified limitations. We further present a strategy suitable for evaluating the low-level characteristics of the generated lateral offset profiles relevant for validating an automated driving function such as a cut-in detection function within simulations. The model’s capabilities are demonstrated based on five single driver datasets. It is shown that the model allows for efficient calibration, is able to replicate the behavior of these drivers, and is characterized by short runtimes. This makes it suitable for the virtual validation of automated driving functions.","PeriodicalId":100631,"journal":{"name":"IEEE Open Journal of Intelligent Transportation Systems","volume":"5 ","pages":"566-580"},"PeriodicalIF":4.6,"publicationDate":"2024-07-29","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"https://ieeexplore.ieee.org/stamp/stamp.jsp?tp=&arnumber=10612773","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"141872303","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"OA","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Pub Date : 2024-07-22DOI: 10.1109/OJITS.2024.3430843
Praveen Abbaraju;Subrata Kumar Kundu
Electric vehicles (EV) are gaining wide traction and popularity despite the operational range and charging time limitations. Therefore, to ensure the reliability of EVs for realizing improved customer satisfaction, it is necessary to monitor and track its battery condition. This paper introduces a novel federated & ensembled learning (FEL) algorithm for precise estimation of battery State of Health (SoH). FEL algorithm leverages real-world data from diverse stakeholders and geographical factors like traffic and weather data. A Long-Short Term Memory (LSTM) model has been implemented as a base-model for SoH estimation, continuously updating for each trip as an edge scenario using data-centric federated learning strategy. A stacked ensemble learning algorithm is employed to combine data from heterogenous data sources for retraining the base-model. The effectiveness of the proposed FEL algorithm has been evaluated using NASA battery dataset, showing significant improvement in SoH estimations with a mean average error of 3.24% after 30 iterations. Comparative analysis, including LSTM model with and without ensembled stakeholder data, reveals up to 75% accuracy improvement. The proposed model-agnostic FEL algorithm shows its effectiveness in precise SoH estimation through efficient data sharing among stakeholders and could bring significant benefits for realizing data-centric intelligent solutions for connected EVs.
尽管电动汽车（EV）在续航能力和充电时间方面存在限制，但它正日益受到广泛关注和欢迎。因此，为了确保电动汽车的可靠性，提高客户满意度，有必要监控和跟踪其电池状况。本文介绍了一种新颖的联合与集合学习（FEL）算法，用于精确估算电池健康状况（SoH）。FEL 算法利用了来自不同利益相关者和地理因素（如交通和天气数据）的真实世界数据。长短期记忆（LSTM）模型已作为 SoH 估算的基础模型实施，利用以数据为中心的联合学习策略，作为边缘场景对每次行程进行持续更新。采用堆叠集合学习算法，将来自不同数据源的数据结合起来，对基础模型进行再训练。使用 NASA 电池数据集对所提出的 FEL 算法的有效性进行了评估，结果表明，经过 30 次迭代后，SoH 估计有了显著改善，平均误差为 3.24%。对比分析（包括有和无利益相关者数据集合的 LSTM 模型）显示，准确率提高了 75%。所提出的与模型无关的 FEL 算法通过利益相关者之间的高效数据共享，显示了其在精确 SoH 估算方面的有效性，并可为实现以数据为中心的互联电动汽车智能解决方案带来显著效益。
{"title":"A Novel Federated & Ensembled Learning-Based Battery State-of-Health Estimation for Connected Electric Vehicles","authors":"Praveen Abbaraju;Subrata Kumar Kundu","doi":"10.1109/OJITS.2024.3430843","DOIUrl":"10.1109/OJITS.2024.3430843","url":null,"abstract":"Electric vehicles (EV) are gaining wide traction and popularity despite the operational range and charging time limitations. Therefore, to ensure the reliability of EVs for realizing improved customer satisfaction, it is necessary to monitor and track its battery condition. This paper introduces a novel federated & ensembled learning (FEL) algorithm for precise estimation of battery State of Health (SoH). FEL algorithm leverages real-world data from diverse stakeholders and geographical factors like traffic and weather data. A Long-Short Term Memory (LSTM) model has been implemented as a base-model for SoH estimation, continuously updating for each trip as an edge scenario using data-centric federated learning strategy. A stacked ensemble learning algorithm is employed to combine data from heterogenous data sources for retraining the base-model. The effectiveness of the proposed FEL algorithm has been evaluated using NASA battery dataset, showing significant improvement in SoH estimations with a mean average error of 3.24% after 30 iterations. Comparative analysis, including LSTM model with and without ensembled stakeholder data, reveals up to 75% accuracy improvement. The proposed model-agnostic FEL algorithm shows its effectiveness in precise SoH estimation through efficient data sharing among stakeholders and could bring significant benefits for realizing data-centric intelligent solutions for connected EVs.","PeriodicalId":100631,"journal":{"name":"IEEE Open Journal of Intelligent Transportation Systems","volume":"5 ","pages":"445-453"},"PeriodicalIF":4.6,"publicationDate":"2024-07-22","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"https://ieeexplore.ieee.org/stamp/stamp.jsp?tp=&arnumber=10605904","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"141769896","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"OA","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Pub Date : 2024-07-22DOI: 10.1109/OJITS.2024.3432176
Ioannis V. Vondikakis;Ilias E. Panagiotopoulos;George J. Dimitrakopoulos
The state of road surfaces can have a significant impact on vehicle handling, passenger comfort, safety, fuel consumption, and maintenance requirements. For this reason, it is important to analyze road conditions in order to improve traffic safety, optimize fuel efficiency, and provide a smoother travel experience. This research presents a federated learning analysis that brings together edge computing and cloud technology, by identifying various road conditions through a multi-label road surface classification analysis. The presented analysis prioritizes the privacy of road users’ data and leverages the advantages of collective data analysis while building confidence in the system. Multi-label classification is applied in order to capture complexity by assigning multiple relevant labels, thus providing a richer and more detailed understanding of the road conditions. According to the experiments, this approach efficient classifies road surface images, achieving comprehensive coverage even in scenarios where data from certain edges is limited.
{"title":"FedRSC: A Federated Learning Analysis for Multi-Label Road Surface Classifications","authors":"Ioannis V. Vondikakis;Ilias E. Panagiotopoulos;George J. Dimitrakopoulos","doi":"10.1109/OJITS.2024.3432176","DOIUrl":"10.1109/OJITS.2024.3432176","url":null,"abstract":"The state of road surfaces can have a significant impact on vehicle handling, passenger comfort, safety, fuel consumption, and maintenance requirements. For this reason, it is important to analyze road conditions in order to improve traffic safety, optimize fuel efficiency, and provide a smoother travel experience. This research presents a federated learning analysis that brings together edge computing and cloud technology, by identifying various road conditions through a multi-label road surface classification analysis. The presented analysis prioritizes the privacy of road users’ data and leverages the advantages of collective data analysis while building confidence in the system. Multi-label classification is applied in order to capture complexity by assigning multiple relevant labels, thus providing a richer and more detailed understanding of the road conditions. According to the experiments, this approach efficient classifies road surface images, achieving comprehensive coverage even in scenarios where data from certain edges is limited.","PeriodicalId":100631,"journal":{"name":"IEEE Open Journal of Intelligent Transportation Systems","volume":"5 ","pages":"433-444"},"PeriodicalIF":4.6,"publicationDate":"2024-07-22","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"https://ieeexplore.ieee.org/stamp/stamp.jsp?tp=&arnumber=10606293","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"141769897","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"OA","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
京公网安备 11010802042870号
京ICP备2023020795号-1
扫码关注我们
求助内容：
应助结果提醒方式：