Terrain Classification with Markov Random Fields on fused Camera and 3D Laser Range Data

In this paper we consider the problem of interpreting the data of a 3D laser range finder. The surrounding terrain is segmented into a 2D grid where each cell can be an obstacle or negotiable region. A Markov random field models the relationships between neighboring terrain cells and classifies the whole surrounding terrain. This allows us to add context sensitive information to the grid cells where sensor noise or uncertainties could lead to false classification. Camera images provide a perfect complement to the laser range data because they add color and texture features to the point cloud. Therefore camera images are fused with the 3D points and the features from both sensors are considered for classification. We present a novel approach for online terrain classification from fused camera and laser range data by applying a Markov random field. In our experiments we achieved a recall ratio of about 90% for detecting streets and obstacles and prove that our approach is fast enough to be used on an autonomous mobile robot in real time.