Gesture Image Viewer

Gesture‑controlled image viewer that uses MediaPipe Hand Landmarker to detect pinch gestures and zoom an image in real time.

Demo Video

Click the image to see Demo video!!

Overview

gestureImageViewer.py opens a webcam stream, tracks up to two hands, and lets you scale an on‑screen image simply by pinching your thumb and index finger together. A logarithmic mapping makes the zoom feel natural and responsive.

Features

• Real‑time hand tracking with MediaPipe Tasks
• Detects both single‑hand and two‑hand pinch gestures
• Smooth, logarithmic zoom proportional to the distance change between hands
• Landmark overlay window for visual debugging
• Easy‑to‑tweak thresholds and paths at the top of the script

Requirements

Package	Version
Python	3.10 or newer
OpenCV‑Python	>= 4.8
MediaPipe	>= 0.10
NumPy	latest

Install everything with:

pip install mediapipe opencv-python numpy

Getting Started

1. Place the image you want to view in media/ (default is baboon.png).
2. Download hand_landmarker.task from the MediaPipe release page and put it in mediapipeModel/.
3. Make sure your webcam is connected (or change video_path).
4. Run:

python gestureImageViewer.py

The script will open two windows:

• LIVE_STREAM video – your webcam feed with hand landmarks drawn.
• test – the image that zooms in/out as you pinch.

Controls

Action	Gesture
Start zoom (two‑hand)	Pinch thumb & index fingertips on both hands
Zoom in/out	Move hands apart / together while pinched
End zoom	Release pinch on either hand
Single‑hand zoom	Pinch with one hand; move hand toward/away from camera

Press Esc to quit, or Space to pause/resume the video.

How It Works

1. Hand Tracking – MediaPipe extracts 21 landmarks per hand at each frame.

2. Pinch Detection – is_pinch() measures the normalised thumb‑to‑index distance and compares it with pinch_threshold (0.8–1.0 works well).

3. Scaling – scale_img_w_diag() computes the diagonal distance between the two pinched hands at the start (l_i) and current frame (l_c) and applies:

   scale_factor = math.log(l_c / l_i, 2)

4. Rendering – draw_landmarks_on_image() overlays landmarks, and OpenCV shows both the annotated video and the scaled image.

Customisation

Parameter	Location	Purpose
image_path	top of script	Source image file
model_path	top of script	MediaPipe model location
video_path	top of script	Webcam index or video file
pinch_threshold	main loop	Pinch sensitivity
INTER_NEAREST	cv.resize	Interpolation method

Project Structure

project/
├── gestureImageViewer.py
├── mediapipeModel/
│   └── hand_landmarker.task
└── media/
    └── baboon.png

Troubleshooting

• Laggy video – Lower your webcam resolution or comment out landmark drawing.
• False pinches – Increase pinch_threshold slightly.
• No hands detected – Check lighting; MediaPipe needs clear contrast to pick up fingers.

License

The code is licensed under the GNU AGPL v3.0.

For more details, see the LICENSE file or visit GNU AGPL v3.0 License