Automatic inventory identification through image recognition

Abstract

Companies are looking to use technology to replace repetitive, low skill tasks currently performed by human employees. One such series of tasks is conducting an inventory of items within a warehouse. This requires employees to identify individual items on a shelf. This task is well-suited to the application of computer vision techniques. Images of shelves in a warehouse can be acquired and analyzed. The goal of this research is to accurately identify items within the images for the purpose of inventory management. This objective requires both the detection of objects within an image, and the identification of those objects. I propose a series of techniques to identify an item against a set of template images of potential items. To demonstrate the effectiveness of these techniques, I have developed a prototype system that performs an inventory of equipment in a data centre.