Our core technology is powered by neural networks, a game-changing innovation in Computer Vision that excels in detecting objects within shelf images, and classifying these images with unparalleled accuracy. Modeled on the brain’s visual cortex, our deep learning architecture trains on vast amounts of shelf image data to recognize fine-grained differences between objects, overcoming challenges like poor light, reflection, background clutter and partial obstruction.
Since images are taken from multiple viewpoints, we’ve designed a stitching engine that mosiacks shelf images and projects them on to a single reference frame. This visual reconstruction produces an optimal 3D structure that solves the problem of duplicate or removed products, and represents the actual store shelf.
What goes in?
Fully stitched shelf images with correctly identified SKU information
Under the hood
Location and size of each SKU is identified, and this data is used to automatically generate scene and session facts.
What comes out
Digitized shelf data is combined with survey responses to generate numeric measurements, scores and targets.
Our experts have developed a system that uses geometric techniques to account for the angle and distance from the shelf from which the photo was taken. This results in a digital coordinate database that captures spatial details like size and location of each product on every shelf, even accounting for stacked items. For e.g. the Coca-Cola 600 ml bottle was in bay 4, shelf 2, fourth item from right, at the top of the stack. Using this data, scene-level and session-level calculations like shelf share and OSA are performed.
Data discovery and visualization
We collect everything so that you can dive in anywhere and explore and enrich the data the way you want.
Visualize the data with engaging dashboards for those lightbulb moments.
The Trax Difference
We combine cutting-edge technology and a robust quality assurance discipline to recognise any product at a brand and sub-brand level with market-leading accuracy.
Graphics Processing Unit Technology accelerates our image recognition, deep learning, and analytics applications, enabling us to go from image to insight rapidly.
Our machine-learning algorithms and engines detect anomalies like incomplete shelf capture and also spot new SKUs, new design versions to give you a complete view of the market.
Our infrastructure is cloud-agnostic, available at an amazing 99.95% and allows us to process 7 million images a month
Our reporting and data discovery platform gives you flexibility to choose from a wide array of KPI’s, and get role-based dashboards, both on web or mobile.