In this project we shall investigate the benefit of combining four pertinent sensor candidates:
(Stereo) Vision, which is the natural and popular technique that is currently being employed and developed at CLAAS Agrosystems. As stereo data is co-registered with color data, it is possible to label the obstacle detected on the basis of range information with the terrain type estimated by color analysis. Therefore, by combining geometric information obtained by stereo triangulation with terrain classification based on color, some types of obstacle detection can be effectively performed in an outdoor environment. Stereo vision alone cannot however detect all types of obstacles that are encountered in the field; in practice this heavily limits the use of autonomous vehicles in arable farming. CLAAS Agrocom, University of Salento and the Danish Technological Institute have experience with developing dedicated stereo vision algorithms.
Radar, which is able to see through the crop - this was used by nearly all of the successful participants of the DARPA Grand and Urban Challenge events, which required unmanned vehicles to autonomously drive through either desert or urban courses respectively. The strength of this sensor-modality is in particular to detect objects that move and to estimate their velocity. In contrast to optical sensors like vision or ladar, radar technology is capable to see through clouds, smoke and even walls. Cemagref has experience with developing radar solutions.
Ladar, which allows fast 3D perception - is today often used to generate 3D point clouds which represent a model of the environment. Modern Laser range finders cannot only measure the distance to the points, but also their reflectance. Hence, one gets a greyscale image, where the distance of each pixel is known, similar to the result of a so-called time-of-flight camera, but with a significant larger possible distance between sensor and object. These images offer rich opportunities to apply and improve algorithms for object detection. Fraunhofer has experience with developing ladar solutions.
Thermography, which can detect humans and animals, independent of fog and lighting conditions. This technology is today offered by high-end car manufactures in brands such as BMW, Audi and Rolls Royce. The system improves night vision and vision in foggy or dusty conditions considerably. The Danish Technological Institute is currently investigating the feasibility of applying thermographic cameras in other projects. For instance for animal observation in a farming environment. In the later mentioned project cameras are mounted on a pan/tilt unit on top of a mobile surveillance robot. An advantage of adapting thermography in agricultural vehicles is that the technology is currently becoming cheaper likely due to the fact that the technology seems to be moving towards mass production.
None of the above mentioned four sensors will be able to solve the problem alone, but they will deliver data, that have to be fused in order to maximize the fidelity of the vehicle's ambient awareness.