ESA title


  • ACTIVITYKick-start Activity
  • STATUSCompleted
  • THEMATIC AREAInfrastructure & Smart Cities

Objectives of the service

Image credit:

Within the retail landscape, the distribution of products in stores is in constant change. The supply and demand of consumer products in stores is locally very volatile and difficult to predict. Producers and distributors take great efforts to stay informed about the current distribution of their products. The ProDiRec AI service provides the possibility to recognize product distribution in retail stores by analysing crowdsourced retail shelf photos and detecting products on these photos automatically.

ProDiRec service uses precisely geo-mapped photos of retail shelfs to extract information about product distribution and out of stock situations. The geolocated retail shelf photos are gathered using the crowdsourcing platform. By applying AI technology for object detection, the ProDiRec service recognizes predefined products automatically. This reduces the manual effort to a quality control of the results.

From the recognized products on the geolocated photos of retail shelfs, ProDiRec offers regional insights into several KPIs of product distribution. The automated product recognition offers the opportunity to scale the product distribution and avoid out of stock situations of ProDiRec customers.

Users and their needs

The target market includes product manufacturers and distributers. In order to sell their respective products, it is important for ProDiRec customers that their products are continuously present and well displayed in retail stores as intended. Most of ProDiRec customers try to directly or indirectly influence the product distribution in retail stores to improve their revenues. However, they struggle to measure their performance in this endeavour and manually measuring the product distribution does not allow to scale customers’ projects for larger regions or towards a higher number of products. The main needs of ProDiRec customers are:

  • Distribution Detection: Are my products in the store / shelf?
  • “Facings” count: How good are my products positioned in the store / shelf?
  • Repeatability: How does my retail presence develop over time?
  • Cost-effective large-scale projects: How to reduce the manual effort for detecting products?
  • Historic data is not available: How to cost-effectively recognize products on existing images?

All of the needs can be addressed by collecting information via crowdsourcing area-wide in each end consumer store by crowdworking platforms such as However, targeting for the analysis of such a massive amount of current or historical image data requires automatic solutions as provided by ProDiRec.

Service/ system concept

The ProDiRec service depends on three main building blocks: The crowdsourcing service, which enables the gathering of retail shelf photos required for the analysis. The backend, where the crowdsourcing jobs are derived from the project definition of ProDiRec customers. And the product detection, where the analysis of the photos is performed in order to derive the product distribution data.

An integral part of the service is to gather reliable data from the retail stores. Therefore, ProDiRec relies on  appJobber crowdsourcing service to retrieve the geolocated retail shelf photos. The crowdsourcing jobs for every point of interest (i.e. retail store) are published through the appJobber app, where the jobbers of the crowd can find them on a map to reserve and complete them. The SatNav system allows to reliably pinpoint a single photo to geocoordinates, which allows a spatial analysis of product distribution later.

The product recognition part works iteratively, where data analysts manually annotate a number of products first. These annotations are used to train an object detection model, which recognises products and reduces the manual annotation to quality control. Recognising further products on photos in ProDiRec database allows to provide product distribution data to the customers.

Image credit:

Space Added Value

A very important aspect is to gather reliable data using the crowdsourcing service. Using appJobber crowdsourcing app, the jobbers can find their jobs on a map to reserve and complete them. Taking photos outside and inside of retail stores within dense city centres requires a reliable measurement of geolocation to prove the origin of the retail shelf photos and attribute the correct store to the shelf photo. Therefore, the crowdsourcing service depends on Satellite Navigation with EGNOS correction data for improving the accuracy on a level of house number accuracy. This approach was result of Business Incubation Project ESOC 17-2010 „Participatory Sensing“ and is described more in detail in the corresponding reports.

mage credit:

Current Status

Image credit :

Through the kick start, a standard process for image attribution of FMCG products was developed, confirming the technical feasibility of AI-based automated product recognition.

The economic viability demonstrated the relevance of ProDiRec to enable the execution of large-scale projects in a cost-effective way.

At the end of the kick start, formal confirmations from two pilot customers to support ProDiRec product development and pilots were received.

A follow-up demonstration project is planned to further develop the solution and pilot it in operational settings.

Image credit:

Status Date

Updated: 22 May 2020