Detecting events by actors using dynamically cropped images

Abstract

Systems within materials handling facilities or retail establishments are programmed to receive images from cameras, process clips of the images to generate sets of features representing product spaces and actors depicted within such images, and to classify the clips as depicting or not depicting a shopping event. The images are dynamically cropped to reduce amounts of data that must be processed to in order to determine whether the images depict shopping events. The images are cropped by calculating a center point based on positions of points on product spaces and detected overlaps of hands and the product spaces. Features of clips determined to depict a shopping event are combined into a sequence and transferred, along with classifications of such clips, to a multi-camera system that generates a shopping hypothesis based on such features.