Multiple-sensor based approach for road terrain classification

Publication Type:
Issue Date:
Full metadata record
Files in This Item:
Filename Description Size
Thumbnail01front.pdf200.76 kB
Adobe PDF
Thumbnail02whole.pdf3.96 MB
Adobe PDF
Complete perception of the environment around a vehicle plays a crucial role in safe driving. Those include other stationary or dynamic objects, drivable regions, road signs and terrain types. In this thesis, classification of terrain types is investigated. Knowledge of the road terrain is useful to improve passengers’ safety and comfort in road vehicles. It is also useful for bounding safe navigational routes for autonomous vehicles. Therefore, the aim of this thesis is to develop a methodology to identify impending road terrain types by using on-board sensors. Two kinds of sensors are used in this research. The first kind of sensors measure parameters of the vehicle in terms of vibration and speed. Therefore, these sensors can only produce measurements while the vehicle is navigating on a particular terrain. The second type of sensors measure properties of the terrain in terms of structure and visual cues, for example cameras and Laser Range Finders (LRFs). These sensors can even produce measurements of impending terrain types. However, all those kinds of sensors have their own advantages and disadvantages. In this thesis, it is proposed to fuse them to improve the terrain type classification results. The sensor fusion is achieved using Markov Random Field (MRF). The MRF model is designed to contain five nodes and five cliques which describe the relationships between the classification results of the accelerometer, camera, and two LRFs. The MRF model’s energy function is appropriately synthesized to improve the classification accuracies of the impending terrains. Experiments carried out on a real vehicle test-bed, CRUISE (CAS Research Ute for Intelligence, Safety and Exploration) on different types of roads with various speeds show that the MRF based fusion algorithm lead to significant improvements (approximate 30%) of the road terrain classification accuracies.
Please use this identifier to cite or link to this item: