Thoht Delta
HomeServicesOur WorkBlogCareers Get in Touch
All Posts
Engineering August 28, 2024

Computer Vision for Retail: It Goes Way Beyond Counting Heads

Raja Siddharthan

When most retailers hear “computer vision,” they think of one thing: counting how many people walk through the door. And yes, footfall counting is valuable. But it’s like buying a smartphone just to make phone calls — you’re missing 90% of what the technology can do.

At Thoht Delta, we’ve built CV systems that go deep into retail operations. Here’s a broader view of what’s possible when you point AI at a camera feed.

Beyond Counting: What CV Can Actually Do in Retail

Customer Journey Mapping

Footfall tells you how many people entered. Journey mapping tells you what they did inside. By tracking anonymized movement paths through the store, you can answer questions like:

  • Which entrance do most customers use?
  • What’s the most common path through the store?
  • Where do customers pause, and where do they walk past without looking?
  • How does the journey differ between customers who buy and those who leave empty-handed?

This data is gold for store layout optimization. Instead of guessing which product should go where, you place products along the paths your customers actually walk.

Shelf and Display Monitoring

One of the most expensive problems in retail is out-of-stock situations. A customer walks up to a shelf, finds the product missing, and either buys a competitor’s product or leaves. The retailer loses the sale and may not even know it happened.

CV can monitor shelves in real time:

  • Detect empty shelf spaces and trigger restocking alerts
  • Verify that promotional displays are set up correctly
  • Check planogram compliance — is the product where it should be?
  • Measure which displays get the most visual attention

Queue Management

Long checkout queues are the number one reason customers abandon purchases. CV can:

  • Count people in queue in real time
  • Estimate wait times based on current throughput
  • Alert staff to open additional counters when queues exceed thresholds
  • Track queue patterns over time to optimize staffing schedules

Demographic-Driven Merchandising

When you know the age and gender distribution of your visitors (anonymously, from camera analysis), you can make smarter merchandising decisions:

  • A store that discovers 60% of its afternoon visitors are women aged 25-40 can adjust product displays, music, and promotions accordingly
  • A store near a college that sees mostly 18-25 year olds on weekends can stock and display differently than on weekday mornings
  • Seasonal demographic shifts become visible in data instead of remaining invisible hunches

Security and Loss Prevention

Traditional CCTV security is reactive — you watch the tape after something happens. CV-powered security is proactive:

  • Detect unusual behavior patterns (loitering, repeated visits without purchase)
  • Track items being picked up but not reaching the checkout
  • Monitor restricted areas and alert when unauthorized access occurs
  • Identify when the store is unusually empty or crowded for the time of day

The Technology Stack

Making all of this work requires several layers:

  • Detection models — identifying and classifying people and objects in each frame
  • Tracking algorithms — maintaining identity across frames as people move through the store
  • Re-identification — recognizing the same person across different camera views
  • Edge compute — processing video locally to avoid bandwidth costs and latency
  • Analytics platform — turning raw detections into actionable dashboards and alerts

Each layer has its own engineering challenges. The detection model needs to work across varying lighting conditions. The tracker needs to handle occlusion when people walk behind shelves. Re-identification needs to work even when someone takes off their jacket between cameras.

What’s Coming Next

The next wave of retail CV will likely include:

  • Emotion and engagement detection — understanding not just where customers go, but how they feel about what they see
  • Product interaction tracking — knowing which items were picked up, examined, and put back vs. purchased
  • Predictive staffing — using historical traffic patterns to forecast staffing needs days or weeks in advance
  • Cross-store analytics — comparing customer behavior across hundreds of locations to identify what top-performing stores do differently

Getting Started

You don’t need to implement everything at once. Most retailers start with footfall counting and demographics — the foundational layer. Once that data starts informing decisions, the appetite for deeper analytics grows naturally.

The infrastructure investment is minimal if you already have CCTV cameras. An edge compute box and a cloud dashboard get you started. More advanced features can be layered on as you scale.

Curious about what CV can do for your stores? Talk to us — we’ll help you figure out what to measure first.

computer visionretail technologyAIstore operations

Related Posts

Want to learn more?

We build computer vision systems for retail, transit, and industrial applications.

Get in Touch