Combining Structure from Motion with visual SLAM for the Katwijk Beach dataset

On November 26th, 2015 a planetary rover drove across the Katwijk beach.
Man-made rocks were arranged on the beach to imitate a Martian landscape. While driving the rover recorded several datasets. These can be used for the application of different SLAM techniques.

In this study the focus is on a single sensor in the datasets: the stereo camera. Three SLAM techniques are applied to the stereo camera footage, to get a good estimate of the position of the rover and a reliable map. These techniques are: visual SLAM using points clouds, Structure from Motion, and a combination of both techniques. Using the point clouds from visual SLAM and the location estimations from Structure from Motion, this study was able to create a visually more comprehensible map with a more accurate location estimation of the rover. The results indicate that by combining the techniques a better performance on both mapping and localisation can be achieved.