Updated AI Shelf Recognition

August 19, 2021

Image Recognition (IR) technology has seen great advances over the past few years and allowed field sales personnel much more time to focus on their core roles of selling and merchandising and reduced time spent on data collection, price logging and in-store admin.

The IR technology developed by SoftServe Business Systems (SSBS) is called AI Shelf Recognition and has seen some great successes in leading edge FMCG businesses, such as Reckitt. In their business case they saw cost reductions of 30% and work acceleration in store of 30%.

This type of outstanding result is not unusual and is the result of a dedicated team of professional developers and programmers continuing to stretch boundaries by improving technology speed, quality delivery and ease of use, all whilst reducing the cost to serve.

Over the last five years SSBS have been engaged in leading these changes and seen their system evolve through the following stages:

  1. Opencv Feature Detection (SIFT and SURF) – this technology found function zones and compared them with a reference image, which took several days to prepare, and about 10 minutes to recognize the image in store. The cost was very high and accuracy debatable, however it did raise awareness that the technology worked and was available.
  2. Neural Networks Phase 1– this required fewer resources for recognition (R-CNN). R-CNN could be trained, but the learning process required thousands of photos and many man-hours to tag each photo. Here preparation to be system ready took 4-8 weeks and 10-30 seconds for recognition of the image in store. Price started to reduce and become more affordable as the technology improved.
  3. Neural Network (NN) + Augmentation Phase 2– Augmented Reality has reduced the NN learning process as well as recognition time and cost. It only requires 2 weeks for preparation, 10-30 seconds for recognition, and the cost has become much lower.
  4. Neural Networks Phase 3 – a new generation of neural networks now allows recognition by a mobile device in 5-20 seconds. Thanks to the optimization of the internal processes of the service, the training of the neural network now only takes 1-2 weeks. There is a much lower cost of service, high quality images are taken and image recognition is consistently high.

Due to SSBS’s dedication to a twin approach of improving IR technology and reducing the cost to our customers all the old ‘pain points’ have disappeared. The technology now features AI, offline working, ultra-fast data collection and real-time interface with HQ systems and users. And if that wasn’t enough our AI Shelf Recognition can now recognise price tags in store, competitor products, POS placements, promotional displays, and secondary placement.

If you would like more information about the technology or have some questions please feel free to contact Vladimir Kvyatkovsky at vkvyatk@softserveinc.com.