Vision For Retail - V4RETAIL

Start Date and End Date

07 January 2015
01 January 2016

Turkish Partner(s)

Vispera Bilgi Teknolojileri Sanayi İç ve Dış Ticaret A.Ş.


Vispera Bilgi Teknolojileri Sanayi İç ve Dış Ticaret A.Ş.


50.000 Euro


Horizon 2020 SME Instrument

Project Web Page

Scientific Outputs

V4RETAIL, the automated product recognition and counting system developed by Vispera will increase the efficiency of the auditing operations and minimize human errors. This will lead to improvement of the business processes of the FMCG companies and  result in economic gain.

Tackling ‘grand’ or societal challenges

One of the major pain points in retail is the lack of speed and precision in information from the selling floor. Acquiring accurate and real-time feedback on inventory distortions, such as out-of-stock, overstock and misplaced items is of paramount importance for all retail operations. The current practice of supermarket shelf monitoring and auditing relies on the visual inspection of supermarket shelves by a human auditor. This visual inspection mainly involves recognizing packaged products displayed on the shelf and producing an approximate shelf share value for each. 

Industrial Innovation (including innovation in services as well as products and processes)

The solution consists of an end-to-end software service operating on a series of mobile, desktop and cloud-based software applications. In particular, the vision technology improves the state-of-the-art in product recognition, which is mostly limited to operate in “one visible item per image (or per processed patch)” basis and lacks efficiency when multiple instances from multiple categories are present in cluttered scenes. In that aspect, the contribution of the project to the related scientific field is multi-fold: 

Given the large number of SKUs and complexity of shelf scenes, our approach is very efficient thanks to hierarchical visual dictionaries that project use for matching image features to their precomputed product counterparts. Second, project process shelf scenes as a whole by reasoning about the common scene attributes holistically to eliminate spurious matches, and resolving distinct detections in parallel with Generalized Hough transform from feature matches to the SKU-pose space. Third, project exploit the fact that shelf items are arranged under a planogram, and accordingly combine our results with spatial scene priors in a Bayesian sense to achieve an improved and context-aware visual recognition system.

Research-influenced changes in policy, agenda-setting

The provision of Improved Public Goods

The improved exercise of professional skill

Wholesaler firms improve their business processes and achieve economic gains

Human capital development