Computer vision

What is computer vision in retail?

Computer vision is a branch of artificial intelligence that trains computers to interpret visual information — images and video — the way humans do. In retail, this means using cameras to automatically understand what’s happening inside a store: how many people are visiting, where they go, how long they stay, and how they interact with products and displays.

Unlike traditional analytics that rely on point-of-sale data or manual surveys, computer vision provides a continuous, objective view of the entire customer experience from the moment someone walks through the door.

Why it matters for retailers

Physical retailers face a fundamental data gap compared to their online counterparts. E-commerce platforms automatically track every click, scroll, hover, and purchase. Until recently, brick-and-mortar stores could only measure what happened at the till.

Computer vision closes this gap by providing the in-store equivalent of web analytics:

• Visitor counting — how many people entered, like unique visitors on a website
• Path tracking — how customers navigate the store, like clickstream analysis
• Dwell analysis — how long people spend in each area, like time-on-page
• Conversion measurement — what percentage of visitors actually bought, like checkout conversion rate
• Demographic insights — anonymous age and gender estimation of your audience

How computer vision works in-store

The technology stack

A typical retail computer vision system consists of:

1. Cameras — standard CCTV or IP cameras already installed in the store
2. Edge processing device — a small on-premise unit that runs the AI models (e.g., Aura Vision’s Analytics Processing Unit)
3. Deep learning models — neural networks trained to detect people, track movement, and classify behaviour
4. Analytics platform — a dashboard or API where insights are visualised and acted upon

What the AI actually does

The deep learning models are trained on millions of examples to perform several tasks simultaneously:

• Person detection — identifying each individual in the frame, even in crowded scenes
• Tracking — following each person’s path across frames and across multiple camera views
• Pose estimation — understanding body orientation and movement patterns
• Demographic estimation — inferring approximate age group and gender from body shape, gait, and clothing (not from faces)
• Staff segmentation — learning the visual appearance of staff uniforms to separate employees from customers

All of this processing happens in real time, typically at the edge (on-device), so that no video footage needs to leave the building.

Key applications

Footfall and traffic analysis

The most fundamental application: counting how many people enter and exit the store with high accuracy, across all entrances, while filtering out staff.

Store layout optimisation

By generating heatmaps and path maps, retailers can see which areas of the store attract the most attention and which are under-visited. This informs product placement, fixture positioning, and overall store design.

Queue management

Computer vision can detect queue formation, measure wait times, and alert staff when lines exceed a threshold — reducing checkout abandonment and improving customer satisfaction.

Campaign and display measurement

Retailers can use A/B testing to measure the impact of different window displays, signage, or promotional setups by comparing footfall and engagement metrics before and after changes.

Demographic profiling

Anonymous age and gender estimation helps retailers understand who their customers are across different stores, times of day, and seasons — without collecting any personal data.

Privacy considerations

Privacy is the most important consideration when deploying computer vision in retail. The best systems are designed with privacy by design:

• Edge processing — video is analysed on-device; no footage is transmitted or stored in the cloud
• Face blurring — applied at source with non-reversible algorithms before any image could theoretically leave the device
• No personal data — outputs are always anonymous aggregates (counts, percentages, heatmaps)
• Compliance — leading systems are GDPR, CCPA, and EU AI Act compliant, with certifications like SOC 2 Type II

How Aura Vision uses computer vision

Aura Vision’s approach connects a compact edge device (the APU) to existing CCTV cameras. Deep learning models — trained in-situ to adapt to each store’s specific cameras, lighting, and environment — run entirely on the device. The system outputs anonymous analytics including footfall counts, heatmaps, path maps, dwell times, demographics, and queue metrics.

No new cameras are needed, installation takes the same day, and the entire system operates without any video leaving the building.

Frequently asked questions

Does computer vision in retail use facial recognition?

Not necessarily. Privacy-conscious systems like Aura Vision explicitly avoid facial recognition. Instead, they use body shape, gait, and clothing analysis to understand demographics and track customer journeys — all without identifying individuals.

Do you need special cameras for computer vision?

No. Modern retail computer vision systems are designed to work with standard CCTV or IP cameras that are already installed in most stores. There is no need for specialised sensors or new hardware.

How does computer vision protect customer privacy?

Privacy-first systems process video entirely on-device (edge processing), apply non-reversible face blurring at source, and only output anonymous aggregate data such as counts and heatmaps. No video footage or personally identifiable information leaves the premises.

What is the ROI of computer vision in retail?

Retailers typically see ROI through improved conversion rates (by optimising layouts and staffing), reduced queue abandonment, and more effective marketing campaigns. Many retailers report measurable uplift within the first few months of deployment.