On-Shelf Load Cell Calibration for Positioning and Weighing Assisted by Activity Detection: Smart Store Scenario

Accurate tracking of shelves in markets is necessary for efficient and real-time management of stock and products. Smart technologies assist in determining the positions of items and the movement to or from the shelves. Recently, several shelf tracking systems have been proposed utilizing sensor fusion, image processing, deep learning, and the Internet of Things. However, these systems suffer from the imbalance calibration of load cell sensors, high cost of hardware installation, and low positioning accuracy. Therefore, we propose an accurate online Weighing Distribution Positional Model that enables accurate load cell calibration and tracks item position and weight on shelves utilizing the load cells sensor fusion and deep learning. The load cell sensors are used to weigh shelves, and cameras are used to record the item’s movement which is analyzed by image processing. Four well-known machine learning algorithms are tested on the system to identify the best operative environment. The system and the algorithms are evaluated using a real-world dataset consisting of item weights, and the Artificial Neural Network (ANN) was found more accurate than the others with 97.61% accuracy and zero error tolerance.