Graph-based modelling of superpixels for automatic identification of empty shelves in supermarkets

Bikash Santra, Udita Ghosh, Dipti Prasad Mukherjee

Abstract: Highlights•A graph-based model of superpixels is introduced for segmenting gap/non-gap regions in the shelf images of supermarkets.•A shelf image is over-segmented into superpixels to create a graph of superpixels (SG).•The nodes and edges of a SG are uniquely encoded using our graph convolutional and Siamese networks.•A structural support vector machine is formulated with the SG for finding gaps in shelves.•Our annotations for three retail product datasets labelling gap/non-gap regions are released at https://github.com/gapDetection/gapDetectionDatasets.