[Paper Review] Global 6DOF Pose Estimation from Untextured 2D City Models

Clemens Arth, Christian Pirchheim+2 more|arXiv (Cornell University)|Mar 9, 2015

Robotics and Sensor-Based Localization15 references3 citations

TL;DR

This paper presents a method for global 6DOF camera pose estimation in outdoor urban environments using only untextured 2D city maps and inertial sensor data. It estimates absolute camera orientation from straight line segments and refines translation by aligning the 2D map with semantic segmentation of the input image, achieving accurate, absolute pose estimation without requiring textured images or multiple registered views.

ABSTRACT

We propose a method for estimating the 3D pose for the camera of a mobile device in outdoor conditions, using only an untextured 2D model. Previous methods compute only a relative pose using a SLAM algorithm, or require many registered images, which are cumbersome to acquire. By contrast, our method returns an accurate, absolute camera pose in an absolute referential using simple 2D+height maps, which are broadly available, to refine a first estimate of the pose provided by the device's sensors. We show how to first estimate the camera absolute orientation from straight line segments, and then how to estimate the translation by aligning the 2D map with a semantic segmentation of the input image. We demonstrate the robustness and accuracy of our approach on a challenging dataset.

Motivation & Objective

• To enable accurate, absolute 6DOF camera pose estimation in outdoor urban scenes without relying on textured images or multiple registered views.
• To address the limitation of SLAM-based methods that only provide relative pose estimates.
• To leverage widely available 2D+height city maps as a global reference for pose refinement.
• To integrate inertial sensor data with 2D map alignment for robust and precise pose estimation.

Proposed method

• Estimate the camera's absolute orientation using detected straight line segments in the input image.
• Use a 2D+height city map as a global geometric reference for pose refinement.
• Align the 2D map with a semantic segmentation of the input image to estimate camera translation.
• Fuse inertial sensor data with map-based constraints to improve pose accuracy.
• Refine the initial sensor-based pose estimate using a joint optimization framework over map and image features.
• Leverage the geometric consistency between 2D map contours and image line segments to constrain the solution space.

Experimental results

Research questions

• RQ1 Can absolute 6DOF camera pose be estimated in outdoor urban environments using only untextured 2D city maps?
• RQ2 How can inertial sensor data be effectively combined with 2D map geometry to improve pose accuracy?
• RQ3 Can line segment detection and semantic segmentation be used to estimate orientation and translation independently and then fused?
• RQ4 What level of accuracy and robustness can be achieved without textured images or multiple registered views?
• RQ5 How does the proposed method compare to SLAM-based approaches in terms of absolute pose accuracy?

Key findings

• The method achieves accurate global 6DOF pose estimation using only untextured 2D city maps and inertial sensor data.
• Orientation estimation via line segments provides a robust initial estimate even in feature-sparse environments.
• Translation refinement through 2D map and semantic segmentation alignment significantly improves pose accuracy.
• The approach outperforms relative pose methods by providing absolute pose in a global coordinate system.
• The method is robust on a challenging dataset, demonstrating practical applicability in real-world urban settings.
• The use of 2D+height maps enables precise localization without requiring textured or high-resolution image data.