Connected vehicle networks generate massive volumes of heterogeneous sensor data that require efficient processing for real-time traffic management applications. Traditional centralized approaches face significant challenges including privacy concerns, communication overhead, and scalability limitations. This paper presents a novel framework that integrates federated learning with multi-sensor fusion and edge computing optimization to address these challenges. The proposed system employs a distributed architecture where connected vehicles collaboratively train machine learning models while preserving data privacy through differential privacy mechanisms and homomorphic encryption. Our edge computing optimization algorithm dynamically allocates computational resources based on traffic density, sensor data quality, and network conditions. The multi-sensor fusion component integrates data from imaging sensors, LiDAR, ultrasonic sensors, and DSRC communications to enhance decision accuracy. Experimental validation using real-world traffic datasets demonstrates that our approach achieves 23.7% improvement in traffic flow efficiency compared to centralized systems while reducing communication overhead by 45.2%. Traffic flow efficiency is defined as <InlineEquation ID="IEq1"> <EquationSource Format="TEX">\(\:EFE=\left({V}_{avg}/{V}_{max}\right)*\left(1-D/{D}_{max}\right)\)</EquationSource> </InlineEquation>, where <InlineEquation ID="IEq2"> <EquationSource Format="TEX">\(\:{V}_{avg}\)</EquationSource> </InlineEquation> is average vehicle speed, <InlineEquation ID="IEq3"> <EquationSource Format="TEX">\(\:{V}_{max}\)</EquationSource> </InlineEquation> is maximum speed limit, <InlineEquation ID="IEq4"> <EquationSource Format="TEX">\(\:D\)</EquationSource> </InlineEquation> is traffic density (vehicles/km), and <InlineEquation ID="IEq5"> <EquationSource Format="TEX">\(\:{D}_{max}\)</EquationSource> </InlineEquation> is jam density. This is measured via simulation on MTC datasets, with improvement calculated against a centralized baseline where EFE=0.62 (ours: 0.767). This 23.7% improvement is statistically significant (t-test: t = 4.56, p < 0.001, 95% CI [18.2%, 29.2%]) over 10-fold cross-validation. The framework maintains privacy guarantees with epsilon-differential privacy of 0.1 and achieves real-time processing with average latency under 50 milliseconds. Results indicate significant potential for enhancing intelligent transportation infrastructure while addressing critical privacy and scalability concerns in modern connected vehicle deployments. Graphical Abstract Figure 1 illustrates the graphical abstract of our proposed privacy-preserving federated multisensor fusion framework with edge optimization for real-time traffic management

Privacy-Preserving Federated Multi-Sensor Fusion with Edge Computing Optimization for Real-Time Traffic Management in Connected Vehicle Networks

期刊论文
发表日期：

2025年11月27日

查看全文

Milad Rahmati,
Nima Rahmati

摘要

Connected vehicle networks generate massive volumes of heterogeneous sensor data that require efficient processing for real-time traffic management applications. Traditional centralized approaches face significant challenges including privacy concerns, communication overhead, and scalability limitations. This paper presents a novel framework that integrates federated learning with multi-sensor fusion and edge computing optimization to address these challenges. The proposed system employs a distributed architecture where connected vehicles collaboratively train machine learning models while preserving data privacy through differential privacy mechanisms and homomorphic encryption. Our edge computing optimization algorithm dynamically allocates computational resources based on traffic density, sensor data quality, and network conditions. The multi-sensor fusion component integrates data from imaging sensors, LiDAR, ultrasonic sensors, and DSRC communications to enhance decision accuracy. Experimental validation using real-world traffic datasets demonstrates that our approach achieves 23.7% improvement in traffic flow efficiency compared to centralized systems while reducing communication overhead by 45.2%. Traffic flow efficiency is defined as \(\:EFE=\left({V}_{avg}/{V}_{max}\right)*\left(1-D/{D}_{max}\right)\) , where \(\:{V}_{avg}\) is average vehicle speed, \(\:{V}_{max}\) is maximum speed limit, \(\:D\) is traffic density (vehicles/km), and \(\:{D}_{max}\) is jam density. This is measured via simulation on MTC datasets, with improvement calculated against a centralized baseline where EFE=0.62 (ours: 0.767). This 23.7% improvement is statistically significant (t-test: t = 4.56, p < 0.001, 95% CI [18.2%, 29.2%]) over 10-fold cross-validation. The framework maintains privacy guarantees with epsilon-differential privacy of 0.1 and achieves real-time processing with average latency under 50 milliseconds. Results indicate significant potential for enhancing intelligent transportation infrastructure while addressing critical privacy and scalability concerns in modern connected vehicle deployments.

Graphical Abstract

Figure 1 illustrates the graphical abstract of our proposed privacy-preserving federated multisensor fusion framework with edge optimization for real-time traffic management