Your POS knows exactly how many receipts a store printed yesterday. It has no idea how many people walked in and left without buying. That missing number is why two stores with the same sales can have very different problems: one may lack visitors, while the other is losing the visitors it already has.
People counting fills that gap. Cameras or sensors at the entrance count who comes in and out, and heatmaps show where visitors spend time inside the store. Combined with transaction data, they let you calculate conversion rate, match staff schedules to real traffic and test whether a new layout or promotion changed behavior.
Accuracy depends mostly on the counting method and camera placement. Privacy depends on design choices you make before installing anything. This guide covers both, plus how to turn the numbers into decisions.
What people counting and heatmaps measure
| Metric | How it is calculated | Decisions it supports |
|---|---|---|
| Footfall | Entries crossing a virtual line at the door, per hour or day | Staffing, opening hours, campaign impact |
| Conversion rate | Transactions divided by visitors in the same period | Sales effectiveness, staff coaching |
| Capture rate | Entries divided by people passing the storefront | Window displays, signage, location value |
| Dwell time per zone | Time visitors remain inside a defined floor area | Product placement, promo zones |
| Occupancy heatmap | Visitor density across the floor over time | Layout changes, aisle congestion |
| Queue length and wait | People inside the cashier queue zone | When to open another counter |
One nuance matters in Indonesia: people often shop in groups. A family of four visiting a mall store on Sunday adds four entries but usually one transaction, so raw conversion looks low on weekends. Some systems estimate groups from how closely people enter together. At minimum, compare weekday and weekend conversion separately rather than as one average.
Choosing a counting method
| Method | Strengths | Limitations | Best fit |
|---|---|---|---|
| Overhead 3D or time-of-flight sensor | High entrance accuracy, handles crowds, usually no identifiable image | Covers the door only, needs ceiling mounting and one unit per entrance | Malls, flagship stores, busy entrances |
| CCTV video analytics | Reuses cameras, adds heatmaps, queues and zones | Sensitive to angle, lighting and occlusion; needs compute | Stores that want floor analytics, not just counts |
| Thermal sensor | No identifiable image, works in low light | Less detail, fewer analytics | Privacy-sensitive locations |
| Infrared beam | Low cost, simple | Miscounts people walking side by side, limited direction data | Small shops wanting a basic count |
| Wi-Fi or Bluetooth signals | Covers wide areas, hints at repeat visits | Modern phones randomize Wi-Fi MAC addresses, so counts are rough | Trend indicators only |
For most retail chains, the practical choice is between dedicated entrance sensors and CCTV analytics. Sensors give a dependable door count with little tuning. CCTV analytics takes more effort to set up but answers more questions, from heatmaps to queues, and can reuse cameras you already own. Many stores combine them: a sensor at the door for the official count, cameras inside for behavior. Our article on retail CCTV analytics covers the wider set of in-store use cases.
How CCTV-based people counting works
A video pipeline for counting has four stages:
- Detection. A model finds each person in every frame. Our guide to object detection explains how these models are trained and deployed
- Tracking. A tracker links detections across frames, so each person keeps one temporary ID while they move
- Rules. Virtual lines and zones turn tracks into events: an entry when a track crosses the door line inward, a dwell event when it stays in a zone longer than a set time
- Aggregation. Events are summed into 15-minute or hourly counts and sent to a dashboard, usually without storing the video itself
Camera placement decides accuracy
At the entrance, a camera mounted overhead or at a steep angle sees people from above, so they rarely block each other. A camera at head height facing across the door will miss people walking behind others. Cover the full width of the entrance, and avoid pointing the camera at glass doors with strong daylight behind them, a common issue for street-facing shophouse (ruko) stores.
Where counts go wrong
- Staff counted as customers. Exclude staff with a separate staff entrance, uniform recognition or by subtracting known staff movements
- Lingering at the door. People who stop in the doorway or step in and out can be double counted unless the rules require a full crossing
- Children and strollers. Decide whether children count as visitors, then test that rule
- Changing light. Evening light, reflections and infrared night mode all change what the model sees
For heatmaps, the camera's perspective distorts distance, so the image must be mapped to a floor plan. Calibrate each camera against known points on the floor, or the heatmap will exaggerate activity near the camera.
Validate accuracy before you trust the numbers
Do not accept a vendor's accuracy figure without testing it in your own stores. A simple validation process:
- Record sample video at different times: a quiet weekday morning, a weekend peak and an evening under artificial light. If your pilot runs during Ramadan, include the hours after iftar, when shopping patterns often change
- Have two people count entries and exits manually from the recordings, then resolve their differences
- Compare system counts with the manual counts for entries and exits separately
- Agree on acceptance thresholds per entrance in your contract, before rollout
- Re-validate after any layout change, camera move or firmware update
A system that is slightly but consistently low is still useful for trends. A system whose error swings between busy and quiet hours will mislead every comparison you make, so check accuracy at peak times, not only on average.
Privacy: count people, not identities
Video of shoppers can be personal data under Indonesia's Personal Data Protection Law (Law No. 27 of 2022, or UU PDP). Counting does not need to know who anyone is, so design the system that way:
- Process video on-site or on edge devices, and send only counts and anonymous track data to the cloud
- Do not use face recognition for traffic counting. It adds risk without improving the count. The trade-offs differ for uses like attendance, which we cover in face recognition for attendance.
- Keep raw footage only as long as your security policy requires, and restrict who can view it
- Put up clear notices that the area is monitored by cameras, and state the purpose
- Sign data processing agreements with any vendor that handles footage
UU PDP requires a data protection impact assessment for processing with high potential risk, and systematic monitoring can fall into that category. Ask your legal team whether your setup qualifies; our guide to UU PDP compliance covers the wider obligations.
Turning foot traffic into decisions
Counts become valuable when they sit next to sales and staffing data. Useful questions to answer first:
- Staffing. Do cashier and floor staff schedules match hourly traffic, especially on weekends and around end-of-month paydays?
- Conversion. Which hours have high traffic but low conversion? That pattern often points to staffing, stock or queue problems worth investigating
- Campaigns. Did a promotion bring more visitors, or did it convert the same visitors better? Compare against similar stores that did not run it
- Layout. After moving a category, did dwell time in that zone change?
- Store comparison. Conversion is a fairer way to compare stores of different sizes than sales alone
Put these into a daily view for store managers and a weekly view for regional heads, rather than another report nobody opens. Our guide to KPI dashboards for executives shows how to decide what each level sees. For more retail use cases, visit our retail industry page.
How Weaveward can build this for you
We would start with a site survey of one or two pilot stores: entrance layout, existing CCTV, lighting and network. Then we set up counting on entrance cameras or sensors, add zone analytics only where they answer a real question, and connect the counts to your POS data in a store dashboard. Before rollout, we validate the counts against manual tallies from your own footage and agree on accuracy targets with you.
See how we approach computer vision for retail and operations, or talk to our team about a foot traffic pilot.