Stereo vision with texture learning for fault-tolerant automatic baling

Research highlights▶ Robust vision using texture analysis for tracking of field structures. ▶ Fault-tolerant learning using stereo vision and texture. ▶ Automatic baling obtained using vision and feedback from implement. ▶ Field tests validated the quality of theoretical results.AbstractThis paper presents advances in automated baling using stereo vision. A robust classification scheme is developed for learning and classifying based on texture and shape. Using a state-of-the-art texton approach a fast classifier is suggested that can handle non-linearities and artifacts in data. Shape information is employed to make the classifier robust to large variations in lighting conditions and greatly reduce the likelihood that artifacts in signals from the stereo vision system lead to gross errors in estimated object positions. The classifier is tested on data from a stereovision guidance system on a tractor. The system is shown to be able to classify cut plant material (called swath) by learning its appearance. A 3D classifier is successfully used to train the texture classifier. It is demonstrated from field tests how fault-tolerant fusion of steering reference data are obtained for an automated baling vehicle.