Combined road prediction and target tracking in collision avoidance

Eidehall, A.; Gustafsson, F. · 2004 · Crossref

DOI: 10.1109/ivs.2004.1336455

archive: archived pipeline: cataloged verified

Get this paper ↗ (DOI — opens at the source; we link to it, we don't host it)

Summary

This paper addresses the challenge of estimating lane geometry and tracking other vehicles simultaneously for intelligent driver assistance systems, such as collision avoidance and adaptive cruise control. The authors argue that integrating these two estimation tasks into a single filter allows for more optimal utilization of available information. Specifically, the motion of tracked vehicles can improve lane curvature estimates during poor visibility, while accurate road geometry aids in interpreting vehicle motion. The core technical problem involves handling the non-linearities introduced by mapping curved road coordinates to Cartesian coordinates within a Kalman filter framework. To solve this, the authors derive a coordinate transformation that maps positions from a curved coordinate system following the road to a Cartesian system attached to the host vehicle. They model road curvature using a clothoid curve, where curvature changes linearly with distance. Since the exact transformation involves non-integrable trigonometric functions, the paper derives and evaluates three specific approximations: (A) omitting the clothoid parameter to assume constant curvature, (B) linearizing trigonometric functions, and (C) further simplifying the linearized model. These approximations are incorporated into an Extended Kalman Filter (EKF) state space model that includes host vehicle states (lane width, offset, heading, curvature) and tracked vehicle states (position, velocity). The motion model assumes tracked vehicles follow their lanes, simplifying lateral dynamics. The methods were evaluated using real-world data from a vehicle equipped with a camera, radar, and inertial sensors. The authors compared the three integrated filter approximations against a decoupled model where lane geometry and obstacle tracking were performed separately. Filter tuning was conducted by scaling process and measurement noise covariances to optimize curvature estimation accuracy. Results showed that during bad visibility, the integrated filters significantly outperformed the decoupled model, as the system could rely on vehicle motion to correct lane estimates when visual data was unreliable. In good visibility conditions, the improvement was marginal. Additionally, the study examined lane assignment accuracy, demonstrating that precise curvature estimation is critical for correctly identifying the lanes of distant vehicles, where small heading errors result in significant lateral deviations. The significance of this work lies in demonstrating the practical benefits of coupled estimation for automotive safety systems. By integrating road geometry and target tracking, the system becomes more robust to sensor degradation, such as poor visibility, which is a common failure mode for vision-based lane detectors. The paper provides specific mathematical approximations that make this integration computationally feasible for real-time applications, offering a pathway to more reliable collision avoidance and lane guidance systems.

Provenance

The full processing record for this entry. Every stage of this paper's journey through the pipeline is logged — what ran, with which tool and model, how many attempts it took, and when it last completed.

Summary generated by qwen3.6-27b-prismaquant on 2026-06-26; verification: verified.

Topics

Ranked by relevance to this paper. Hover a topic for its definition.

• anticipation