.png){kind=icon}
.png){kind=icon}
Return
Digital twin intersection based on roadside multi-sensor data fusion
.png){kind=icon}
PRE
AI
DOI：10.48130/dts-0025-0023.png)
.png){kind=icon}
Original request for help.png){kind=icon}
Share.png){kind=icon}
Save
Abstract
En 中文
Digital twin technology is pivotal in the advancement of smart cities and autonomous driving due to its unique capabilities in virtual-reality integration, interactive control, and predictive analysis. The primary enabler for achieving advanced transportation digital twins lies in enhancing environmental sensing capabilities, with multi-sensor data fusion emerging as a widely adopted strategy to improve sensing performance. However, existing research has predominantly focused on onboard systems, leaving roadside sensor deployment and roadside multi-sensor data fusion strategies insufficiently explored. Recognizing the potential advantages of roadside sensor systems, such as broader sensory field coverage and reduced occlusion. This study investigates the integration of roadside multi-sensor data fusion with digital twin technology in the transportation domain. Consequently, this paper introduces an innovative intersection digital twin system developed through a simulation-based approach, leveraging roadside multi-sensor data late fusion. The Car Learning to Act-Simulation of Urban Mobility (CARLA-SUMO) co-simulator acts as a data generation platform, synchronously producing RGB images and Light Detection and Ranging (LiDAR) point clouds with spatiotemporal consistency. For object detection, we employ the You Only Look Once version 5 (YOLOv5) and PointPillars algorithms. Then, a decision-level fusion strategy is proposed to integrate these heterogeneous sensor outputs into a cohesive roadside digital twin system. Experimental results demonstrate that YOLOv5 and PointPillars achieve a mean Average Precision (mAP) of approximately 90% and 60%, respectively. Moreover, the detection frequency of both detectors is well-suited to the dynamic nature of intersection traffic, and the fusion strategy synergistically exploits the complementary advantages of heterogeneous sensors to enhance overall system performance. This research contributes to the field by facilitating low-cost autonomous driving simulation tests and enabling the reconstruction of intersections using roadside digital twin technology, with significant implications for vehicle-road coordination and traffic management.
Keywords:
Digital twin
CARLA-SUMO co-simulator
Multi-sensor data fusion
YOLOv5
PointPillars
Journal
D
IF:
0
Papers:
15
Citations:
0