Skip to main content

Paper Published: Road Geometry Estimation Using Vehicle Trails

· 2 min read
Yi-Chen Zhang
Yi-Chen Zhang
Lead Engineer, AI and Autonomous

I am pleased to share that my paper titled "Road Geometry Estimation Using Vehicle Trails: A Linear Mixed Model Approach" has been published in the Journal of Intelligent Transportation Systems: Technology, Planning, and Operations.

Overview​

The paper addresses a fundamental problem in autonomous highway driving: accurately estimating the shape of the road ahead using motion trajectories of leading vehicles.

A vehicle trail is essentially the historical path of a vehicle, where position samples are collected longitudinally from the same vehicle over time. Such measurements are naturally available from sensor fusion systems — whether from a single sensor or multi-sensor tracking setups. The key idea is to exploit these trails from leading vehicles to infer the geometry of the road in highway scenarios.

Method​

The proposed method is built around a polynomial-based road model estimated via a Linear Mixed Model (LMM) — one of the most widely used frameworks in statistical modeling. The approach addresses two practical engineering challenges:

  • Memory efficiency: Before feeding trail data into the LMM framework, the data undergoes a newly developed compression and chopping mechanism to prevent memory overload from accumulating trail samples.
  • Computational efficiency: The profile likelihood function is used to reduce the number of iterations in the Newton-Raphson optimization, significantly alleviating the computational burden.

Results​

The method was evaluated on two publicly available NGSIM (Next Generation Simulation) datasets. Large-scale simulation results show that the road shape estimated by the proposed method achieves an RMSE of less than 0.5 meters on average across all evaluation ranges when compared with the ground truth road shape — demonstrating both accuracy and robustness in capturing road geometry.