Automating Inventory Management with Computer Vision Techniques

Abstract

Efficient storage management is critical in many industrial settings. This thesis explores novel ways to apply machine learning and computer vision techniques to automate the process, whilst focussing on making this technology affordable and accessible enough for adoption by small and medium sized enterprises.

To showcase the possibility of computer vision used in inventory management, a prototype system was developed using a webcam overlooking a set of drawers, utilising both deep learning and traditional machine learning techniques to show how this might be possible. A novel bounding box algorithm was developed to avoid the challenge of manual data annotation, resulting in a system that is able to identify open drawers with high accuracy, and assess whether the drawer is empty or if stock is low.

This research presents a practical, low cost method for automated storage monitoring, reducing manual effort and offering potential applications in inventory tracking, robotics, and smart storage solutions. Future work may explore integrating pre-trained models or refining the segmentation approach to enhance classification accuracy further.