Hand Tracking

Real-time hand tracking using MediaPipe and OpenCV. Detects hands via webcam and processes 21 hand landmarks per hand in real time.

Demo

Feature	Status
Webcam capture	✅
Hand detection	✅
Landmark processing	✅
Live display window	✅

How It Works

1. Webcam frame captured via OpenCV (cv2.VideoCapture)
2. Frame converted BGR → RGB for MediaPipe
3. MediaPipe Hands model detects up to 2 hands and returns 21 3D landmarks per hand
4. Processed frame rendered in an OpenCV window

MediaPipe Hand Landmarks

MediaPipe tracks 21 keypoints per hand:

  8   12  16  20
  |   |   |   |
  7   11  15  19
  |   |   |   |
  6   10  14  18
  |   |   |   |
  5   9   13  17
   \  |   |  /
    4 |   | 4
     \|   |/
      3   3
      |   |
      2   2
      |   |
      1   1
       \ /
        0  (WRIST)

Landmark indices map to: 0 = Wrist, 1-4 = Thumb, 5-8 = Index, 9-12 = Middle, 13-16 = Ring, 17-20 = Pinky.

Requirements

• Python 3.11+
• Webcam / camera device
• uv (recommended) or pip

Installation

Install as package via pip

pip install git+https://github.com/muhammadsufiyanbaig/hand-tracking.git

Install as package via uv

uv add git+https://github.com/muhammadsufiyanbaig/hand-tracking.git

Local development with uv (Recommended)

git clone https://github.com/muhammadsufiyanbaig/hand-tracking.git
cd hand-tracking

uv sync

Local development with pip

git clone https://github.com/muhammadsufiyanbaig/hand-tracking.git
cd hand-tracking

python -m venv venv

# Windows
venv\Scripts\activate

# macOS / Linux
source venv/bin/activate

pip install -e .

Usage

CLI (after package install)

hand-tracking

Module

python -m hand_tracking

Script (local dev)

# uv
uv run main.py

# activated venv
python main.py

Press Q to quit.

Project Structure

hand-tracking/
├── hand_tracking/
│   ├── __init__.py         # Package init, exposes __version__
│   └── __main__.py         # Core tracking logic + CLI entry point
├── main.py                 # Thin shim for direct script execution
├── hand_landmarker.task    # MediaPipe hand landmark model (task file)
├── pyproject.toml          # Package metadata, dependencies, build config
├── uv.lock                 # Locked dependency versions
├── .python-version         # Pinned Python version (3.11)
└── README.md

Dependencies

Package	Version	Purpose
mediapipe	0.10.9	Hand landmark detection model
opencv-python	≥4.13.0	Webcam capture and image rendering

Configuration

mpHands.Hands() accepts optional parameters:

Parameter	Default	Description
static_image_mode	False	True for images, False for video stream
max_num_hands	2	Maximum number of hands to detect
min_detection_confidence	0.5	Minimum score to consider detection valid
min_tracking_confidence	0.5	Minimum score to keep tracking (video mode)

Example with custom config:

hands = mpHands.Hands(
    max_num_hands=1,
    min_detection_confidence=0.7,
    min_tracking_confidence=0.6
)

Extending the Project

Common next steps:

Draw landmarks on frame:

mpDraw = mp.solutions.drawing_utils
if results.multi_hand_landmarks:
    for handLms in results.multi_hand_landmarks:
        mpDraw.draw_landmarks(img, handLms, mpHands.HAND_CONNECTIONS)

Read a specific landmark (e.g. index fingertip = landmark 8):

if results.multi_hand_landmarks:
    for handLms in results.multi_hand_landmarks:
        h, w, _ = img.shape
        lm = handLms.landmark[8]
        cx, cy = int(lm.x * w), int(lm.y * h)

Display FPS:

pTime = 0
# inside loop:
cTime = time.time()
fps = 1 / (cTime - pTime)
pTime = cTime
cv2.putText(img, f"FPS: {int(fps)}", (10, 70),
            cv2.FONT_HERSHEY_PLAIN, 3, (255, 0, 255), 3)

License

MIT