Reversing Shop view Analysis for Planogram Creation

• Main Author: Barbosa, Telmo Domiciano Pereira
• Format: Dissertation
• Language: English
• Published: ProQuest Dissertations & Theses 01-01-2017
• Subjects: Algorithms; Automation; Biology; Cameras; Compliance; Computer science; Marketing; Morphology; Product placement; Retail stores; Software
• ISBN: 9798835555932

With the increasing care of retail shop owners in improving sales and costumer experience, there is a need to develop technology in order to optimize their goals. It’s proven that a planned product placement can boost sales and improve costumer experience [CSDK07]. With this in mind, Fraunhofer Portugal came up with ShopView [RGC+16] solution to help retail shops extract, validate and manipulate planograms from high-resolution images of the real shelves in the store.In this sense, this thesis focused on the creation of an algorithm using computer vision algorithms to extract information from high resolution images of retail shelves taken with the ShopView solution. Particularly, it was implemented pre-processing steps to improve the efficiency and accuracy of an OCR (Optical Character Recognition) engine in recognizing the text in the shelves products. These pre-processing algorithms comprise of denoising and segmentation techniques. The use of this OCR engine brings additional information about the products, this information is later used in clustering algorithms to automatically extract an accurate planogram from shelves photos.The presented algorithm is capable of extracting relevant information from the shelves images, to identify the existing products and create a valid metadata about them and their location. With this metadata, it is possible to create, validate and modify the planogram in ShopView. The use of OCR on this algorithm has advantages over other available approaches due to its capability to differentiate products with minimal visual differences and its immunity to appearance changes on the products packaging. Moreover, the methodology proposed does not require any previous user interaction to work properly.