Advanced Particle Filter-Based V2V Collaborative Localisation in Urban Environments

Background: Urban environments pose significant challenges for autonomous vehicle localisation due to dynamic conditions, lack of reliable GNSS position information and limited map-based guidance provided by LiDAR scan matching. To address this limitation, a collaborative localisation approach is proposed, leveraging a stereo camera setup to detect respective platform calibrated markers on the rear of leading vehicles. This cooperative strategy enables the enhancement of positional accuracy – particularly for the longitudinal direction - by integrating data from multiple traffic agents. An advanced particle filter is employed to fuse sensor inputs, ensuring robust and reliable localisation in complex urban scenarios.

Objective: This master thesis aims to develop and integrate a reliable Vehicle-to-Vehicle (V2V) collaboration approach into an existing Particle Filter framework. By incorporating the uncertainty information of leading vehicles, the method seeks to accurately quantify the filtered ego-vehicle's pose. The primary objective is to minimize longitudinal deviations, particularly in feature-poor environments, where current methods experience significant pose disturbances. Ultimately, this module aims to enhance localization robustness and accuracy in challenging urban scenarios.

Specific Methodology:

Data Preparation:

• Filter misdetections in real-world image sequences.
• Estimate the stochastic model of the marker detection algorithm.
• Format data for compatibility with the particle filter framework.

Collaborative Implementation:

• Develop and experimentally validate a stereo camera-based collaboration approach.
• Incorporate uncertainty information from aiding vehicles.
• Integrate LiDAR scans for long-distance support if beneficial.

Evaluation:

• Perform Monte Carlo simulations to assess performance.
• Run a standard filter for the aiding vehicle to estimate poses.
• Use aiding vehicle poses for ego-vehicle collaborative localization and analyse results

Programming Requirement:

MATLAB: Proficiency is essential for integrating the collaboration module into the particle filter framework.

Python: Required for image sequence analysis, implementing collaboration concepts, and conducting extensive evaluations.