Efficient Multi-Sensor Fusion for Cooperative Autonomous Vehicles Leveraging C-ITS Infrastructure and Machine...

Area of Science:

• Transportation Engineering
• Computer Science
• Signal Processing

Background:

• Cooperative Intelligent Transport Systems (C-ITS) require robust sensor fusion for real-time data processing.
• Infrastructure sensors in C-ITS often produce noisy and asynchronous data.
• Existing fusion strategies face challenges in handling data variability and computational load.

Purpose of the Study:

• To propose a novel two-stage data fusion framework for C-ITS applications.
• To improve the accuracy and efficiency of sensor data fusion in dynamic environments.
• To address the challenges of noisy, asynchronous, and multi-sensor data integration.

Main Methods:

• A two-stage data fusion framework combining a grid-based indexing method and a Light Gradient Boosting Machine (LGBM) augmented by an Extended Kalman Filter (EKF).
• Stage 1: Hybrid EKF-LGBM model for noise mitigation, trajectory refinement, and sensor stream synchronization.
• Stage 2: Grid-based indexing for efficient object association and merging of multi-sensor measurements.

Main Results:

• The proposed framework demonstrates a balance between near-real-time performance and improved trajectory accuracy.
• Outperforms Unscented Kalman Filter (UKF) at a noise scale of 1, achieving 1.81x speed improvement.
• Real-world tests show a 1.54x Root Mean Square Error (RMSE) improvement over baseline measurements.
• Efficiently filters noise and reduces computational overhead for practical C-ITS feasibility.

Conclusions:

• The developed two-stage sensor fusion framework is effective for C-ITS applications.
• The approach significantly enhances data processing capabilities in terms of speed and accuracy.
• The system offers a practical and efficient solution for integrating multi-sensor data in intelligent transport systems.