reCamera detects people on the camera and publishes results over MQTT; a computer or reComputer R1100 runs the dashboard and keeps the history.
Start the MQTT broker, database, Grafana dashboard and video gateway. Every camera publishes here. Deployment discovers ONVIF cameras on the same subnet; if the backend and cameras are on different subnets, enter the camera addresses in the deploy form.
Make sure Docker Desktop is running.
Only needed for cameras ONVIF discovery cannot see — one per entry as id=url, comma separated
| Symptom | Fix |
|---|---|
| Port conflict error | Free ports 8086, 3000, 8080 and 1883 |
| Docker not available | Start Docker Desktop |
| Not enough disk | Keep at least 2 GB free |
Install the retail analytics app on the reCamera and set the broker it publishes to. The app reports dwell state (browsing / engaged / needs assistance) and entry/exit counts.
192.168.42.1; over Ethernet or WiFi, find the IP in your router admin pageThis step disables Node-RED autostart; the two cannot run at the same time.
| Symptom | Fix |
|---|---|
| Cannot connect | USB: use 192.168.42.1; network: check router for IP |
| No data showing | Make sure Step 1 completed and camera and server are on the same network |
Fails to start with a device_init assertion | Reboot the camera |
The heatmap shows the camera's perspective by default. Use the built-in calibration tool to show it on your store's floor plan.
Calibrated cameras share one floor plan. The dropdown at the top-left narrows the view to a single camera.
You only need the camera-view heatmap.
| Symptom | Fix |
|---|---|
| Heatmap doesn't align well | Open settings, click Reset, and recalibrate with different reference points |
| Calibration survives a browser change | Expected; calibration is stored on the server |
Click below to open the Grafana dashboard.
| Symptom | Fix |
|---|---|
| Page not loading | Make sure Step 1 deployed successfully; run docker ps on the server to check services |
| Wrong host/port | If you deployed to a remote device, use its IP in the URL |
| No data | Make sure Step 2 completed and the MQTT server IP is the machine from Step 1 |
admin / admin, traffic trendsreCamera Pro runs detection, tracking, dwell states and entry/exit counting on the device; a computer or reComputer R1100 runs the dashboard.
retail-vision app installed from the reCamera Pro App Center. This solution does not provide its install package.Start the MQTT broker, database, Grafana dashboard and video gateway. Skip if already deployed. Deployment discovers ONVIF cameras on the same subnet; if the backend and cameras are on different subnets, enter the camera addresses in the deploy form.
Only needed for cameras ONVIF discovery cannot see — one per entry as id=url, comma separated
Enter the device's web console credentials (not SSH), an installation name, and the MQTT address of the machine from step 1. With the MQTT address empty, results stay on the camera's page and do not reach the dashboard.
| Symptom | Fix |
|---|---|
| Says retail-vision is not in the App Center | Install it from the device's App Center first, then re-run this step |
| No data on the dashboard | Check the MQTT address points at the machine from step 1 and both are on the same network |
| Avg Dwell tile stays empty | Expected; the Pro app does not report per-person dwell time. Other tiles are unaffected |
Click below to open the Grafana dashboard (login admin / admin).
| Symptom | Fix |
|---|---|
| Page not loading | Make sure Step 1 deployed successfully; run docker ps on the server to check services |
| Wrong host/port | If you deployed to a remote device, use its IP in the URL |
Keep your existing IP cameras; a reComputer RK3588 or RK3576 runs people-flow detection locally.
Start the MQTT broker, database, Grafana dashboard and video gateway. Skip if already deployed. Deployment discovers ONVIF cameras on the same subnet; if the backend and cameras are on different subnets, enter the camera addresses in the deploy form.
Only needed for cameras ONVIF discovery cannot see — one per entry as id=url, comma separated
Deploy the detector to the board over SSH. Pick the right board model; the wrong one will not load the model. If the dashboard runs on this board, leave the MQTT address at 127.0.0.1.
The board needs Docker and the NPU driver (/usr/lib/librknnrt.so present).
rk3588Picks the model built for this NPU — the wrong one will not load
RTSP stream of the camera this board watches
store-01First segment of the MQTT topic — a store, floor or site name keeps several installations apart on one broker
cam-01127.0.0.1Where the dashboard backend runs. Leave at 127.0.0.1 if you deployed it on this same board.
1883| Symptom | Fix |
|---|---|
| Container fails with a librknnrt error | The NPU driver is missing; check /usr/lib/librknnrt.so exists |
| No data on the dashboard | Verify the camera URL with ffprobe rtsp://..., and check the MQTT address |
| Low frame rate with a pinned CPU | Hardware decode is not active; check the MPP libraries on the board |
| Symptom | Fix |
|---|---|
| Page not loading | Make sure Step 1 deployed successfully; run docker ps on the server to check services |
| Wrong host/port | If you deployed to a remote device, use its IP in the URL |
admin / adminKeep your existing IP cameras; a reComputer Industrial R20 with a Hailo-8 accelerator runs people-flow detection locally.
Start the MQTT broker, database, Grafana dashboard and video gateway. Skip if already deployed. Deployment discovers ONVIF cameras on the same subnet; if the backend and cameras are on different subnets, enter the camera addresses in the deploy form.
Only needed for cameras ONVIF discovery cannot see — one per entry as id=url, comma separated
Deploy the detector over SSH. The Hailo runtime version is checked first; a mismatch stops the deployment and shows the installed version.
The board needs Docker and HailoRT 4.21 (driver, user library and GStreamer plugin on the same version).
RTSP stream of the camera this board watches
store-01First segment of the MQTT topic — a store, floor or site name keeps several installations apart on one broker
cam-01127.0.0.1Where the dashboard backend runs. Leave at 127.0.0.1 if you deployed it on this same board.
1883| Symptom | Fix |
|---|---|
HAILO_OUT_OF_PHYSICAL_DEVICES | Another application holds the accelerator; stop it first |
| Deploy reports a libhailort version mismatch | Move the board to HailoRT 4.21, changing driver and user library together |
/dev/hailo0 missing | The accelerator is not seated, or the hailo_pci driver is not loaded |
| No data on the dashboard | Verify the camera URL with ffprobe rtsp://..., and check the MQTT address |
| Symptom | Fix |
|---|---|
| Page not loading | Make sure Step 1 deployed successfully; run docker ps on the server to check services |
| Wrong host/port | If you deployed to a remote device, use its IP in the URL |
admin / adminKeep your existing IP cameras; an NVIDIA Jetson runs people-flow detection locally.
Start the MQTT broker, database, Grafana dashboard and video gateway. Skip if already deployed. Deployment discovers ONVIF cameras on the same subnet; if the backend and cameras are on different subnets, enter the camera addresses in the deploy form.
Only needed for cameras ONVIF discovery cannot see — one per entry as id=url, comma separated
Deploy the detector to the Jetson over SSH. The first deployment builds a TensorRT engine, which takes 2-5 minutes; later deployments reuse it.
The Jetson needs JetPack 6.x, Docker and the NVIDIA runtime.
RTSP stream of the camera this Jetson watches
store-01First segment of the MQTT topic — a store, floor or site name keeps several installations apart on one broker
cam-01127.0.0.1Where the dashboard backend runs. Leave at 127.0.0.1 if you deployed it on this same Jetson.
1883| Symptom | Fix |
|---|---|
| Connection timeout | Check the network, verify the Jetson IP with ping |
| NVIDIA runtime error | Run nvidia-smi on the Jetson to confirm the GPU is available |
| No data | Verify the RTSP URL with ffprobe rtsp://..., and check the MQTT address |
| Slow first start | TensorRT engine build, first time only, 2-5 minutes |
| Symptom | Fix |
|---|---|
| Page not loading | Make sure Step 1 deployed successfully; run docker ps on the server to check services |
| Wrong host/port | If you deployed to a remote device, use its IP in the URL |
admin / admin