An Intelligent Video Analytics Solution to Optimize Retail Operations
The client
Retailers have been collecting video for years — Walmart alone gathers 2.5 petabytes of unstructured data an hour from a million customers — and most of it sits unused.
Netguru set out to find where processing video data actually gets hard, and to build a retail analytics solution general enough to be worth having.
The challengeWhat the teams actually needed
Collecting video is easy. Extracting something from it, transforming it and analysing it is where the difficulty starts.
Challenge 01Everything happening at once
A video frame holds people, objects and movement together, and pulling any one of them out individually is hard.
Challenge 02Expensive at scale
Unstructured data resists being combined into a meaningful insight, and extracting it at volume costs a lot — particularly when done by hand.
Challenge 03Privacy is not optional
Collecting video has to respect privacy law and be anonymised, which constrains the whole approach rather than being a final step.
What we did
Machine learning reading the camera feed in real time
The system automates what would otherwise be manual observation, which raises operational capacity and cuts the time data takes to process.
How many people are in here
The system measures the number of customers in store, which matters more when occupancy is capped.
Length, as it happens
Queue length is measured in real time rather than reconstructed afterwards from till data.
Cashier management
Cashier service management is informed by what the cameras already see, without anyone watching a monitor.
Automation over manual extraction
The approach is deliberately cost-effective, since manual extraction at scale is what stops most retailers using the video they already hold.
More like thisOther places we scaled product teams for growth.
What’s next