Generation of training data for object-sorting work based on a difference between feature amounts of captured object images

This invention describes a computer program that creates training data for systems that automatically sort objects. It works by taking two images: one before objects are sorted and one after. The program then identifies objects in both images and extracts their unique characteristics. It stores these characteristics as training data, specifically noting objects that were present in both images (and are similar), or objects that appeared or disappeared during the sorting process, often on a conveyor belt where a worker performs the sorting.