Smarter Pharmacy Staffing with Lightweight Computer Vision

The client is a pharmacy chain looking for a computer vision system to track customer flow and improve service efficiency. Their main goal is to detect when queues start forming and automatically alert staff if more help is needed at the counter. They also want to tell the difference between quick online order pickups and longer in-person consultations, to better understand service times. The system should work with their existing CRM to identify order types. Most of their stores are small, so one or two cameras are enough. The solution should be lightweight and run on their current infrastructure.

Challenge

Delayed response to sudden customer surges: Staff may be occupied in non-visible areas (e.g., storage or stocking), leading to unattended counters when multiple customers arrive simultaneously, resulting in longer wait times and reduced service efficiency.

Ambiguity in customer intent and location: Without spatial zoning or movement analysis, the system struggles to tell whether a person standing near the counter is waiting in line, seeking help, or simply browsing nearby displays. This limits the accuracy of queue detection and undermines timely staff response.

Limited camera coverage in small spaces: Many pharmacies are small, with few places to install cameras. This makes it hard to set up multiple devices, so the system needs to work well with just one or two cameras and without complex setups.

High variability across 1,000+ store locations: Rolling out the system across a large network means dealing with diverse layouts, ceiling heights, lighting conditions, and camera placements. These differences impact detection accuracy and complicate setup. The solution must adapt reliably across all sites with minimal calibration.

No customer tracking: Since the system uses snapshot-based detection instead of continuous video streams, it can’t follow individual customers over time. As a result, it’s not possible to measure how long each person waits or how long they spend being served.

Privacy and consent limitations: Analyzing customer age or gender through video may not be allowed unless customers have clearly agreed to it. This usually requires linking the video data to loyalty program members who have given permission. Without that, such features might not be usable.

Overcomplication risk: The system should be easy to set up and use on the pharmacy’s existing equipment. Adding complex features like real-time tracking or multi-camera setups could slow things down, require more technical support, and make the solution harder to maintain. It’s better to focus on the most useful features that work well without adding unnecessary complexity.

Solution

1. Object Detection (YOLO-based): The system uses YOLO-based object detection to identify and count people in images captured at set intervals (e.g. every 1–3 minutes), avoiding the need for a continuous video stream and minimizing hardware and network load. By defining specific zones — such as the sales floor or behind the counter — it can reliably distinguish between customers and staff, enabling zone-specific analytics and event-based triggers.
2. Zone-Based Spatial Analysis: The system splits the pharmacy into key areas like the entrance, checkout counter, and product shelves. It uses camera data to see where each person is and assigns them to one of these areas. This helps identify when someone is waiting at the counter, how many people are looking at specific shelves (like cosmetics), and whether a staff member is at the register when needed.
3. Queue Estimation via Spatial Proximity: The system checks how many people are standing near the counter and how close they are to each other. If three or more people are within 1–2 meters of the register, it counts as a queue. This works without tracking each person — just by analyzing individual frames. If a pharmacy has more than one counter, a second camera can be added, but in most cases, one wide-angle camera is enough.
4. Event Triggering and Real-Time Notifications: The system uses images from in-store cameras to identify situations like a customer arriving when no pharmacist is at the counter, or when more than six people are in the store. When this happens, it sends automatic alerts to staff. These alerts are based only on computer vision and zone detection — there’s no need to connect to cash registers or other sales systems.
5. Zone Heatmaps & Engagement Metrics: The system uses computer vision to track how many people spend time in different parts of the store, like cosmetics stands, supplement shelves, or the checkout area. It creates visual heatmaps showing which areas get the most attention during the day. This helps pharmacy managers decide where to place products, adjust displays, or add staff where needed.

Features

People Counting via Object Detection

The system uses object detection to count how many customers are in the store by checking snapshots from the camera taken every 30 to 60 seconds. It doesn’t need a full video stream, which makes it easier to run and less demanding on the equipment.

Zone-Based Presence Detection

The system splits the pharmacy into specific areas like the entrance, sales floor, checkout zone, and staff-only section. It uses camera images to count how many people are in each area. This helps tell the difference between staff and customers and supports features like queue alerts and activity tracking — without needing to identify individuals.

Real-Time Queue Alerts

The system continuously analyzes visual data from designated zones and sends notifications when specific conditions are met:

• No staff at counter: A customer is detected in the store, but no pharmacist is visible in the cashier zone.
• Crowding threshold exceeded: The number of detected customers surpasses a predefined limit, indicating rising demand.
• Queue forming: Multiple people are present near the counter, but not enough staff are nearby to handle the load.

These alerts help ensure timely staff response and reduce customer wait times without needing integration with POS systems.

Foot Traffic Statistics

Using object detection on frames captured every 1–2 minutes, the system calculates the number of customers present in the store at each time interval. These counts are aggregated into hourly and daily reports, highlighting peak periods (e.g. 11:00–13:00) and low-traffic windows.

Online vs. Walk-In Service Analysis

The system integrates with the pharmacy’s CRM to distinguish between online order pickups and walk-in visits. By combining camera-based footfall data with order type records, it helps estimate how much traffic is driven by quick pickups versus in-store consultations — supporting better staffing and scheduling decisions.

Optional: Zone Heatmaps / Product Area Interaction

Using frame-by-frame analysis, the system monitors how many customers spend time in different sections of the store (e.g. cosmetics, supplements, medical devices). It aggregates this data to generate heatmaps showing high-traffic areas over time.

Flexible Camera Setup

The system uses one or two cameras — typically covering the entrance and counter areas — and can work with existing surveillance systems. It captures periodic frames, minimizing hardware changes and making installation simple.

Development Process

1. Initial Assessment and Planning

Deployment begins with a technical audit of each pharmacy’s layout, focusing on:

• Floor dimensions and spatial segmentation
• Fixture placement (shelves, counters, displays)
• Entry points and foot traffic flow
• Existing surveillance infrastructure

Most compact stores (<40 m²) can be covered with a single ceiling-mounted wide-angle IP camera (e.g., 2.8mm lens, 90–110° FOV). For dual-counter or segmented layouts, a second fixed-position camera may be added to monitor pharmacist areas.

Each camera is configured to capture static frames (e.g., JPEG over RTSP) at regular intervals, avoiding real-time video streaming. This minimizes compute load and supports deployment on lightweight edge devices or client-managed servers.

With over 1,000 pharmacy sites, each location introduces distinct camera calibration variables — differing ceiling heights, lighting conditions, lens angles, and floorplans. To support scalable deployment, the solution is designed for flexible zone configuration, camera-agnostic input, and minimal manual tuning

2. Camera Setup

Where possible, the system reuses existing IP-based surveillance infrastructure (RTSP-compatible), reducing installation time and hardware costs.

• Each camera captures static frames at fixed intervals (e.g., every 1–3 minutes), minimizing streaming overhead and enabling deployment on lightweight edge devices.
• A single ceiling-mounted wide-angle camera (2.8–4mm lens, 90–110° FOV) typically covers compact stores. For larger or segmented layouts, a second fixed-position camera is added for pharmacist zone monitoring.

The setup supports flexible configuration across 1,000+ pharmacy sites with varying layouts, ceiling heights, and lighting conditions.

3. Frame Capture and Object Detection

The system uses YOLO-based object detection to process still frames captured from in-store cameras. It detects and counts all visible people, then classifies them as customers or staff based on their location within predefined zones (e.g., cashier area, sales floor, pharmacist station).

4. Zone Definition

The camera view is divided into logical zones (e.g., entrance, cashier/pharmacist station, product display shelves) using the SciForce internal toolkit for zone-based analytics. This toolkit supports:

5. Analytics Layer

The system counts how many people appear in each frame and tracks where they are in the store — such as at the entrance, near the cashier, or in certain product areas like cosmetics or supplements. This data is used to create reports showing how busy the store is at different times of day, which areas get the most attention, and when staff might need to be at the counter.

6. Real-Time Event Triggers

The system analyzes each captured frame using object detection and predefined zones (e.g. entrance, counter area). If a customer is detected at the entrance and no staff are visible behind the counter, it triggers a “no staff present” alert. If more than a set number of customers (e.g. 5) are present in the store or concentrated near the cashier area, it sends a crowding or queue alert. These notifications can be pushed to staff devices (e.g. phone, tablet, POS terminal) to prompt immediate action and reduce customer wait times.

7. Notification System

The notification system operates entirely on computer vision outputs, with no need for POS or CRM integration. Using YOLO-based object detection on periodic frames, it monitors activity in defined zones (e.g., customer area, pharmacist zone) and triggers alerts when certain conditions are met.

Alert conditions include:

• A customer enters but no pharmacist is visible
• Customer count exceeds a set limit
• A queue forms with insufficient staff at the counter

All decisions are based on real-time zone presence. For example, if a customer is detected and no pharmacist is present, an alert is triggered. If multiple customers are detected and only one staff member is visible, a second notification can be sent.

Alerts can be delivered in several ways, depending on the pharmacy’s setup:

• HTTP request to an internal app — the system can trigger alerts directly into existing applications used by staff.
• Custom notifier — we can develop a lightweight application that displays pop-up messages or plays a sound on a designated device.
• Third-party messenger integration — alerts can also be sent via a Telegram bot or similar messaging platform if preferred.

Impact

• 40–60% faster staff response during peak times

Real-time alerts reduce counter downtime by automatically notifying staff when queues form or no pharmacist is visible.

• +30% improvement in identifying peak service hours

Automated footfall tracking provides hourly customer counts, helping optimize schedules without manual observation.

• Up to 15% increase in engagement with key product zones

Heatmap data highlights where customers spend the most time, allowing targeted merchandising in high-traffic areas.