Skip to main content

Extending Face Regonition with MCP

Overview​

This guide explains how to deploy a face-recognition workflow using SenseCAP Watcher (Xiaozhi) together with a reComputer Raspberry Pi system. The Watcher captures images and sends them to the reComputer for face matching against a local database using the internal MCP tool.

Hardware Preparation​

SenseCAP Watcher for XiaoZhireComputer AI R2130-12
Get One Now 🖱️
Get One Now 🖱️

Prerequisites​

Follow this guide to install the reComputer Raspberry Pi system: reComputer R2000 Series Getting Started

Execution Result​

Source Code Repositories:​

Deploy Process​

Step1. Note reComputer's IP address​

  • Connect the reComputer to your network, like router

  • Obtain its IP address, example IP used here: 192.168.24.10

Step2. Flash the firmware onto Watcher​

  • Connect the PC to the Watcher using a Type-C cable at the bottom port

  • Open ESP-IDF 5.5 CMD (version 5.5.1) and navigate to a suitable folder

  • Clone the Watcher firmware repository

git clone -b face_rec_api --single-branch https://github.com/suharvest/xiaozhi-esp32.git
cd xiaozhi-esp32
  • Locate the code section for communication with the reComputer
    • File location : main/boards/sensecap-watcher/sscma_camera.cc (line 769)
std::string face_rec_url = "http://192.168.10.131.80001/recognize"

  • Change the IP address to your reComputer’s actual IP

    std::string face_rec_url = <Your_reComputer_actual_IP>
    • example

  • Compile and flash the firmware

idf.py set-target esp32s3
idf.py menuconfig    # Select SenseCAP Watcher board
idf.py build flash

Step3. SSH into reComputer and Prepare Environment​

  • Connect via SSH (replace username and IP as needed)
ssh <Your_reComputer's username>@<Your_reComputer's host IP>
# For example : ssh [email protected]
note

We recommend using the Remote - SSH extension in VS Code for connecting to your reComputer.

  • Install Docker and set docker user
curl -sSL https://linuxmirrors.cn/docker.sh -o install_docker.sh
sudo bash install_docker.sh
bash <(curl -sSL https://linuxmirrors.cn/docker.sh)

sudo groupadd docker
sudo usermod -aG docker $USER
  • Update system software
sudo apt update && sudo apt full-upgrade
sudo rpi-eeprom-update
  • Install uv
curl -LsSf https://astral.sh/uv/install.sh | sh

Step4. Clone the API and update dependencies​

git clone https://github.com/suharvest/face_rec_api.git
cd face_rec_api/
uv sync
  • Install DKMS, HailoRT, add configuration, and reboot
sudo apt-get install dkms
sudo dpkg --install hailort-4.21.0/hailort_4.21.0_arm64.deb hailort-4.21.0/hailort-pcie-driver_4.21.0_all.deb
echo "options hailo_pci force_desc_page_size=4096" | sudo tee /etc/modprobe.d/hailo_pci.conf
  • Check HailoRT status
sudo reboot
hailortcli scan
hailortcli fw-control identify
  • If you see information similar to the following, it means your driver has been installed successfully

Step5. Add Photos and Start the Face Recognition Service​

  • Place photos in the photos/ folder
  • Generate embeddings
uv run scripts/batch_process.py
  • Give execution permission and start the service
chmod +x start_standalone.sh
./start_standalone.sh

  • The service is now running. The Watcher can communicate with the reComputer to check if a face exists in the database. You can stop the service with Ctrl + C.

  • You can add prompt like : "When a person is detected, use the face recognition tool to check name and confidence: if confidence is greater than 40%, greet them by name; otherwise, give a generic greeting."

Step6. Update the Face Database​

  • After stop the service, add new photos to the photos/ folder and regenerate embeddings and restart service.
uv run scripts/batch_process.py
./start_standalone.sh

Step7. Clear the Face Database​

  • Delete all files in the photos/ folder.

  • Remove the embedding.json file in the data/ folder.

  • After clearing, run the service and you will see the same log

FAQ​

Q: How can I make the Xiaozhi AI call the face recognition API?

A: Simply ask the Watcher a question like, "Check if I am in the face recognition system."
The AI will automatically call the MCP tool to check the face database.

Q: What should I do if the face recognition system cannot upload a photo?

A: Please check that the IP address is correct and that the face recognition service is running properly.

Q:Can I deploy the face recognition system on other devices?

A: Yes. As long as the device runs a Linux-based system and can connect to the network, it can be used for deployment. However, the reComputer series offers better computing performance.

Technical Support​

Loading Comments...