People counting in/out (In/Out Face Counting)
The people counting in/out feature uses face recognition to count the number of unique people passing through the entrance and exit at each location. The system matches faces between the exit and entrance cameras to accurately determine how many people actually entered/exited.
- The system detects faces at the entrance and exit cameras.
- It extracts a recognition feature for each face.
- It compares similarity (cosine similarity) between a person at the exit and the list of people at the entrance.
- When similarity exceeds the configured threshold, the system records them as the same person.
Prerequisites
Hardware
- Cameras placed at the exit and entrance of each location.
- Cameras must capture faces clearly (a suitable angle, sufficient lighting). A 2MP+ camera is recommended, with a front-facing or slightly angled view.
Required configuration
- An ROI has been drawn for each camera — see Configuring ROI zones.
- A gate type (in/out) has been assigned to each camera in the zone configuration.
If the ROI or gate type is missing, count results will be significantly inaccurate: the system picks up people walking past in the hallway, and the in/out person-matching algorithm can't run.
Configuration
- For each entrance/exit camera, open Zone Configuration (see Managing cameras).
- Draw the ROI following Configuring ROI zones.
- Assign a gate-type label to the camera: Entrance camera or Exit camera.
- Save the configuration.
Confirm: check that both cameras (entrance and exit) at the location have an ROI and the correct gate-type label before activating the task.
Viewing people-counting reports
Once the task is running and has data, view it in Generating and exporting reports — the Customer Traffic (aggregate) and Customer Presence (per-customer detail) reports both use data from this feature.
Troubleshooting
Counts are significantly off
| Cause | Fix |
|---|---|
| No ROI configured | Draw an ROI for the camera (required) — see Configuring ROI zones |
| Wrong gate type (in/out) | Recheck the gate-type label in the camera configuration |
| Camera facing into backlight | Reposition the camera or add lighting |
| ROI too wide (picking up hallway traffic) | Narrow the ROI to cover only the doorway |
No people are counted
| Cause | Fix |
|---|---|
| Camera not working | Check the stream connection — see Camera diagnostics |
| Face quality too low | Check lighting and camera angle |
Exits don't match entrances
- Check the face-matching threshold (the system's default value).
- Make sure both the entrance and exit cameras capture faces clearly.
- Check that both cameras have an ROI and are active.
- People wearing a full face covering may not be recognized.
- Low light or backlighting reduces recognition quality.
- Cameras must be positioned to capture faces clearly (not too high, not too far away).
Next steps
- Generating and exporting reports
- FAQs — see Q11–Q14 on people-counting accuracy.