Detecting missing products in commercial refrigerators using convolutional neural networks

Out of stock (OOS) is a problem all stores are facing and it reduces their profit. Standard procedures for solving OOS are mostly manual and not scalable. This paper analyzes and proposes an automated and scalable solution for solving OOS problem inside commercial refrigerators. Small, low resolution cameras are placed inside refrigerators. Images taken with those cameras are analyzed with Faster R-CNN and Single Shot Multibox (SSD) models for object detection. Models were trained using transfer learning and their performances were analyzed and compared. After object detection, K-mean clustering algorithm is used to group objects on same shelves. Distance between objects on the same shelf determines if and where the OOS problem is present.