A Unified Multiple-Target Positioning Framework for Intelligent Connected Vehicles

Zhongyang Xiao, Diange Yang, Fuxi Wen, Kun Jiang

TL;DR

This paper introduces a novel framework that unifies vehicle self-positioning and the localization of other road elements (vehicles, pedestrians, etc.) for intelligent connected vehicles (ICVs). By fusing data from on-board sensors, V2X communication, and high-definition (HD) maps, it achieves centimeter-level accuracy and significantly improves the reliability of perception, even in challenging urban environments.

What Problem Does This Paper Address?

Accurate positioning is critical for the safe and efficient operation of intelligent transportation systems (ITS), especially for future autonomous vehicles. Current systems struggle with two main challenges:

  1. Insufficient Self-Positioning: Individual vehicles often lack the precision needed for high-level automation, especially in environments with poor GPS signals (e.g., urban canyons). Relying solely on on-board sensors also faces limitations in accuracy and cost.
  2. Limited Target Localization: Perceiving and accurately locating other dynamic and static objects in the environment (other vehicles, pedestrians, traffic signs) is crucial for situational awareness. Existing on-board sensors have limited sensing ranges, are prone to occlusions, and face a trade-off between accuracy and cost.

Furthermore, existing approaches often treat vehicular self-positioning and target localization as separate problems, leading to inefficiencies and potential cascading errors.

Methods

The core of this paper is a unified theoretical framework that fuses heterogeneous information sources to address both self-positioning and target localization simultaneously. Key aspects include:

  • Multi-Source Data Fusion: The framework integrates data from:
  • On-board Sensors: Cameras, LiDAR, etc., providing local environmental perception.
  • Vehicle-to-Everything (V2X) Communication: Enabling vehicles to share sensor data and positional information with each other (V2V).
  • High-Definition (HD) Maps: Providing precise, pre-mapped information about static road features.
  • Global Navigation Satellite System (GNSS) / Inertial Navigation System (INS): Providing a baseline position and orientation estimate.
  • Joint Optimization: Instead of separate modules, the framework formulates a single optimization problem to estimate the states (position and orientation) of all connected vehicles and the positions of other targets. This is framed as a maximum likelihood estimation problem, which is then converted into a nonlinear least-squares problem.
  • Factor Graph Interpretation: The problem is visualized and understood using factor graphs, where variable nodes represent the states to be estimated, and factor nodes represent the constraints derived from different measurements (V2V, V2F, V2O, GNSS/INS, HD Map).
  • Levenberg-Marquardt Algorithm: The nonlinear optimization problem is solved using the Levenberg-Marquardt method, a standard iterative technique for solving such problems.
  • Measurement Models: Specific models are defined for each data source (on-board sensors, GNSS/INS, HD map) to represent how measurements are related to the true states.

Datasets

The paper does not introduce new datasets. Instead, it uses a simulated traffic scenario generated by VISSIM, a behavior-based traffic flow simulator. This simulation is designed to mimic realistic urban and suburban driving conditions, including varying densities of vehicles, pedestrians, traffic lights, and signs.

Key Insights and Ideas

  • Unification is Key: Treating self-positioning and target localization as a single, unified problem significantly improves overall accuracy and integrity compared to separate approaches.
  • V2X and HD Maps are Game Changers: The fusion of V2X communication (for cooperative perception) and HD maps provides crucial complementary information that overcomes the limitations of individual on-board sensors, especially in challenging environments.
  • Beyond Line-of-Sight: V2X allows vehicles to "see" what other vehicles see, extending perception beyond their immediate line of sight and field of view, thereby improving target sensing integrity.
  • Low-Cost Sensors Can Be Sufficient: The framework demonstrates that even with low-cost GNSS/INS and on-board sensors, centimeter-level positioning accuracy can be achieved when effectively fused with V2X and HD map data.
  • Constraint Density Matters: The performance is directly correlated with the availability and density of constraints. More connected vehicles and features lead to better positioning results, particularly in sparse environments like suburban areas.

Key Results

  • High Positioning Accuracy: The proposed framework achieves self-positioning accuracy of 0.16 m in urban areas, significantly outperforming previous methods (e.g., 1.79 m from Gloria et al. [19]) and approaching the theoretical Cramér-Rao Lower Bound (CRLB).
  • Improved Target Localization: Relative localization RMSE for surrounding targets is 0.17 m in urban areas, also outperforming existing methods.
  • Enhanced Perception Integrity: The framework captures 90.42% of targets within the demanding space in urban areas, a substantial improvement over on-board sensors alone (41.81%), effectively mitigating occlusion and limited sensing range issues.
  • Adaptability: The framework shows improved performance with an increasing number of connected vehicles and features, demonstrating its adaptability to different scenario configurations.

Limitations and Future Work

  • Data Association Assumption: The current framework relies on an assumption of effective data association (matching sensor measurements to known objects/features). The authors acknowledge that this is a limitation and plan to address it in future work by incorporating association methods directly into the optimization process.
  • Communication Delay: The current model does not explicitly account for communication delays inherent in V2X systems. Future research will focus on formulating the communication delay within the data fusion framework.
  • Error Assumptions: While the framework is flexible, the current analysis simplifies GNSS error to a Gaussian distribution. Extending to other error assumptions is mentioned as a possibility.

Summary updated

Abstract

Future intelligent transport systems depend on the accurate positioning of multiple targets in the road scene, including vehicles and all other moving or static elements. The existing self-positioning capability of individual vehicles remains insufficient. Also, bottlenecks in developing on-board perception systems stymie further improvements in the precision and integrity of positioning targets. Vehicle-to-everything (V2X) communication, which is fast becoming a standard component of intelligent and connected vehicles, renders new sources of information such as dynamically updated high-definition (HD) maps accessible. In this paper, we propose a unified theoretical framework for multiple-target positioning by fusing multi-source heterogeneous information from the on-board sensors and V2X technology of vehicles. Numerical and theoretical studies are conducted to evaluate the performance of the framework proposed. With a low-cost global navigation satellite system (GNSS) coupled with an initial navigation system (INS), on-board sensors, and a normally equipped HD map, the precision of multiple-target positioning attained can meet the requirements of high-level automated vehicles. Meanwhile, the integrity of target sensing is significantly improved by the sharing of sensor information and exploitation of map data. Furthermore, our framework is more adaptable to traffic scenarios when compared with state-of-the-art techniques.
BibTeX
@inproceedings{xiao2019unified,
  author = {Zhongyang Xiao and Diange Yang and Fuxi Wen and Kun Jiang},
  title = {{A Unified Multiple-Target Positioning Framework for Intelligent Connected Vehicles}},
  booktitle = {Sensors},
  year = {2019},
  url = {https://openreview.net/forum?id=2bNkv7qP0q5},
}
Figure