>  Current Research > Pose Estimation

back

Pose Estimation

Goal

 

In order to build a correct 3D model from scans and camera images, it is necessary to determine from where these scans and images were taken. Hence, we have developed techniques to determine the pose of our acquisition truck and its sensors, and to register the acquired laser scans with a global coordinate system. 

 

Our approach utilizes techniques originally developed for navigation of mobile robots and consists of the following steps: 

1. Scan matching between subsequent horizontal scans for obtaining an initial path 

2. Global pose correction with Monte-Carlo-Localization on a Digital Surface Model

In the past, we have also investigated alternative methods for global pose correction:

• Position correction by using digital roadmaps and refinement with aerial photos

• Monte Carlo Localization on aerial photos

  3. Roll correction based on images

 

Results

 

Driven path (left image, red) and ground-based scan points (right image, yellow) are registered with global data such as a DSM.

 

(Click on pictures to enlarge)

 

  back to previous page