Tracking of vehicle trajectory by combining a camera and a laser rangefinder

Goyat, Y.; Chateau, T.; Trassoudaine, L. · 2009 · OpenAlex-citations

DOI: 10.1007/s00138-008-0160-0

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 presents a probabilistic method for tracking vehicle trajectories using a sensor system that fuses data from a color camera and a one-dimensional scanning laser rangefinder. The research is motivated by the need to accurately estimate vehicle trajectories through curves to better understand driver behavior and assist road managers in ensuring network safety. The system is designed for outdoor environments where cameras alone face challenges such as illumination variations, shadows, and mutual occlusions. The authors propose a solution that integrates a kinematic vehicle model into a stochastic framework to handle these complexities. The methodology employs a particle filter, a nonlinear stochastic estimation technique, to recursively estimate the vehicle's state, including position, heading, speed, and steering angle. The vehicle dynamics are modeled using a bicycle kinematic model, which assumes low-speed behavior with negligible lateral forces and constant forward speed. The core innovation lies in the observation model, which combines two likelihood functions: one derived from visual data and another from laser rangefinder data. For the visual component, the authors define an original likelihood function based on the projection of a simplified 3D vehicle model (composed of nested parallelepipeds) onto the image. To compute this efficiently, they utilize a line-based integral image approach that approximates the 3D model projection via its convex hull. Additionally, a multi-source sampling algorithm is introduced to merge observations from the camera and laser intrinsically during the resampling phase of the particle filter. The study validates the proposed method using field data collected from actual sites equipped with the camera-laser sensor system. Vehicle trajectories were measured and compared against ground truth data provided by a kinematic GPS, which offers centimeter-level accuracy. The experimental setup involved cameras placed on a tower to cover different sections of a curve, with the system divided into subsystems of camera-rangefinder pairs. The results demonstrate that the modified particle filter can effectively estimate vehicle states by naturally merging multiple observation sources. The use of the 3D model projection and the efficient likelihood computation allows for robust tracking despite the challenges of outdoor vision, such as varying illumination and shadows. The significance of this work lies in its contribution to Intelligent Transport Systems by providing a robust, real-time capable method for vehicle tracking that overcomes the limitations of single-sensor approaches. By integrating kinematic constraints and fusing visual and laser data within a probabilistic framework, the method offers a reliable means of estimating vehicle trajectories. This approach not only improves tracking accuracy but also provides a foundation for analyzing driver behavior in complex road scenarios, thereby supporting efforts to enhance road safety and traffic management.

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.

• situational awareness