mediapipe-example

This project collects sample scripts using MediaPipe and OpenCV for real-time image analysis such as face detection, Face Mesh, hand tracking, pose detection, and face recognition. They can run on a Raspberry Pi 5 at about 15–25 FPS.

Installation

You should install Python 3 and then install the libraries listed in requirements.txt with the command:

pip install -r requirements.txt

Folder structure

• face-detect.py – face detection script
• face-mesh.py – draw Face Mesh points with FPS counter
• hand-tracking.py – detect and track hands
• pose-detect.py – detect body poses
• face_registration/ – face recognition and attendance system
  • face-recognition.py – face recognition using features from Face Mesh
  • face_recognition_ui.py – GUI for face recognition
  • attendance.py – face recognition for logging employee check‑in/out times
  • faces/ – sample images for building face vectors used for recognition Inside this folder create a subfolder for each person, e.g. faces/person-name/ Add several face images of that person (supports .jpg or .png) before running face-recognition.py If faces/ does not exist, create it first with mkdir faces

Usage

Quick launcher

Run run_demo.py to list available demos and execute them without remembering file paths. Extra arguments after -- are forwarded to the target script.

python run_demo.py list
python run_demo.py run face-detect
python run_demo.py run face-registration -- --threshold 0.92

Direct execution

You can still run the scripts manually if you prefer:

python face-detect.py
python face-mesh.py
python hand-tracking.py
python pose-detect.py
python face_registration/face-recognition.py
python face_registration/face_recognition_ui.py
python face_registration/attendance.py

Before running face_registration/face_recognition_ui.py, install Pillow (for example, with pip install Pillow) and verify that face_registration/face_recognition_processor.py is present in the project.

While running face_registration/face-recognition.py, press n to capture and register a new face. The image will be saved and the face_registration/faces/ folder will be updated automatically.

For face_registration/face-recognition.py you can adjust the cosine similarity threshold used for matching by either setting the environment variable COSINE_THRESHOLD or passing the command-line option --threshold, for example:

COSINE_THRESHOLD=0.9 python face_registration/face-recognition.py
# or
python face_registration/face-recognition.py --threshold 0.9

Press q to close each program's display window.

Notes

These scripts are intended for experimentation or studying MediaPipe. Users can adjust parameters within each script to suit their own tasks.

ไอเดียต่อยอดเพื่อให้โปรเจ็กต์น่าสนใจยิ่งขึ้น

• เพิ่มหน้าจอ Dashboard เล็ก ๆ (Streamlit หรือ Gradio) เพื่อโชว์ผลลัพธ์ของแต่ละเดโมพร้อมวิดีโอตัวอย่าง
• เขียน Dockerfile สำหรับรันบนเครื่องที่ไม่มี Python environment พร้อมใช้งาน
• สร้างโมดูลฝึกโมเดลเฉพาะ (เช่น ท่าทางมือสำหรับสั่งงาน) แล้วนำมาเชื่อมกับ run_demo.py
• บันทึกค่า FPS, อุณหภูมิของ Raspberry Pi และบันทึกเป็นกราฟเพื่อเทียบประสิทธิภาพ
• ทำ workflow สำหรับ CI เช่น GitHub Actions เพื่อตรวจสอบว่าทุกสคริปต์ import ได้โดยไม่ error