Method and apparatus for image processing to avoid counting shelf edge promotional labels when counting product labels

摘要

A method for processing retail store shelf images, comprising: receiving an image containing shelf edge content (eg. labels, barcodes or advertisements); identifying a shelf edge region in the image based on determination of the location and apparent orientation of the shelf edge content; demarcating the shelf edge region. Also disclosed is a method and apparatus for edge detection, comprising: receiving an image with an edge region containing patterned content; finding individual occurrences of the pattern by comparing image sections to a reference image; highlighting each instance with commonly orientated geometric patterns (eg. triangles, rectangles or quadrilaterals) with corners above and below each occurrence; clustering the corners; splitting corners in clusters into two sets corresponding to upper and lower bounds of a shelf edge. Quadrilaterals may be estimated using a geometric model fitting and verification algorithm based on point matches between image and reference image. Upper and lower lines may be fitted to clusters the space in between is identified as the shelf edge. A horizontal kernel bandwidth larger than a vertical kernel bandwidth may be used for clustering. Distance between vertical coordinates of corners and cluster centres may be used for clustering. Clusters may be split into sets based on clockwise point ordering of each quadrilateral. Shelf edge region may be obscured so that shelf edge promotional labels are not counted when counting labels of products on shelves.