Reliable 3D product change detection improves retail shelf monitoring
ShelfChange3D: Object-Level 3D Change Detection for Retail Shelf Monitoring
Computer Vision and Pattern Recognition
Summary
Retail shelves need to be checked to make sure products are in stock and in the right place. The authors created a way to compare two 3D images of shelves taken at different times to spot which products changed and exactly where they are using 3D boxes. They also made a large dataset of synthetic and real shelf images with detailed changes annotated, plus a method called ChangeBox that detects and refines these changes. Their method works better than earlier ones, especially when adapting from fake data to real shelf scenes.
What this means in practice
- •For retail automation teams: Use 3D object-level detection to identify out-of-stock or misplaced products on shelves with accurate 3D bounding boxes for robotic shelf monitoring.$Commercial implications: Enables development of automated retail monitoring products that detect and locate product changes in 3D, improving inventory accuracy and robotic restocking.
- •For warehouse robotics teams: Integrate 3D change detection methods to localize changes in warehouse inventory for improved robotic handling and stock updating.
Authors
Lingyi Zhou, Yunke Wang, Mengyu Zheng, Wenbo Wang, Zijian Wang, Chang Xu
Abstract
Reliable shelf monitoring is an important capability for retail automation, yet existing out-of-stock detection methods mainly operate in image space and lack metric 3D localization for downstream robotic systems. We formulate shelf monitoring as object-level 3D change detection: given two RGB-D observations captured at different times, the goal is to identify changed products and localize each change with a 3D bounding box. To support this task, we introduce ShelfChange3D, comprising 145K synthetic and 5K real-world paired RGB-D observations with object-level 3D change annotations. We further propose ChangeBox, an end-to-end framework that jointly reasons over paired observations and predicts object-level 3D change boxes. To improve localization accuracy, we introduce a geometry-based refinement stage that exploits depth and gravity prior to estimate relative pose and refine predicted boxes. Experiments show that ChangeBox outperforms existing change detection baselines, with further gains from refinement and effective transfer from synthetic to real-world observations.