Bag of visual words - A computer vision method applied to bulk material sorting

Every bulk material sorting machine uses classification: objects are either discarded or accepted into one or more bins. In a classical system, both the features and the classifier are constructed by a vision engineer. While successful in the past, this approach is reaching its limits. More challenging tasks and more complex objects call for automated methods to learn both the classification rules and the features. Pattern recognition methods to do the former are well established, but the latter is still open to debate. In a previous work, we have presented an approach to the sorting of wine berries, which is based on bag of visual words [1]. In this paper, we show that the approach is not limited to this product only, but can be extended to other fields as well. In particular, we consider the task of sorting visually similar pebble stones. The method shows nearly perfect classification out of the box. Both the object descriptors and classification rules are learned from an annotated sample. User interaction was only required to obtain the annotations.