Count entries, exits, and zone occupancy entirely on a micro vision AI module.
Count entries, exits, and zone occupancy entirely on a low-cost micro vision AI module, with zero video leaving the site.
Places requiring low-cost footfall and occupancy counting without expensive industrial PCs or cloud video streaming.
Draw a counting line across the entrance to measure customer entry and exit trends throughout the day.
Define a zone around exhibits to monitor dwell occupancy and optimize booth staffing.
Detect genuine human occupancy in rooms without relying on reservation check-ins.
Monitor directional movement along walkways to detect wrong-way traffic and bottleneck surges.
One module connects via a single USB-C cable. For tilted overhead mounting, set the tracking anchor to the bottom center (feet) to prevent false crossings.
The Grove Vision AI V2 module runs detection, tracking and counting locally, sending counts and preview frames over USB-C serial to your panel or system.
The module runs Swift-YOLO on-device and streams structured data over USB serial to the counting panel or host system.
Benchmark measured on Grove Vision AI Module V2 (Himax WE2, Cortex-M55 + Ethos-U55 NPU) with 240x240 input.
| Metric | Measured Value | Platform |
|---|---|---|
| Throughput | About 13 fps | Grove Vision AI Module V2 (Ethos-U55 NPU) |
| Per-frame inference | 48 ms | Swift-YOLO Nano 192 (INT8 quantized) |
| Person detection accuracy | 92.6% mAP | Indoor benchmark dataset |
| Module power consumption | < 1.0 W | 5V USB-C powered |
| Hardware unit cost | ~$16–26 | >90% savings compared to IPC setups |
Detections and counts are streamed over USB serial in standard SSCMA JSON format.
Check the setup and its product list, then follow the deployment guide.
Select site conditions to generate recommended bundles and guides, or let an AI assistant estimate hardware and deploy for you.
Flash the counting firmware, draw your lines and zones, and watch the counts update live. Optionally plug in a XIAO ESP32-S3 for Wi-Fi and MQTT reporting.
3 steps · ~15min
If set to the bounding box center, the box expands rapidly as a person walks towards the camera, dragging the center point forward across the 2D plane. This movement can cross a nearby counting line and register a false crossing. Setting the anchor to the bottom center ties it to the floor plane, ensuring stable trajectories.
Yes. Severe backlighting behind subjects (such as bright glass doors or windows) can create dark silhouettes. We recommend mounting the camera slightly to the side or angling it away from direct light sources.
The module outputs standard SSCMA JSON strings over USB serial. Any Python script (via pyserial), Node-RED flow, or IoT gateway can read the serial port and consume the JSON stream directly.
Yes. Plug a pre-soldered XIAO ESP32-S3 into the female header on the back of Grove Vision AI V2. Connect your phone to its setup hotspot (192.168.4.1) to configure your 2.4GHz Wi-Fi and MQTT broker. The device will then run standalone and publish crossing events directly over MQTT.
Counting events and per-frame detections over USB serial (SSCMA JSON)
Optional wireless MQTT reporting via XIAO ESP32-S3 Wi-Fi add-on (JSON line crossings and zone occupancy)
Grove Vision AI Module V2 connected over USB-C