Object Size Measurement

Using your webcam to a real-time digital ruler using any ID card as a reference.

How it works

1. Place your ID card on the left side of the camera's view
2. The script detects the card by its known aspect ratio (85.6 × 54 mm)
3. It computes a Pixels-Per-Millimetre (PPM) ratio from the card's pixel width
4. Every other shape in the frame is measured using that ratio
5. Dimensions are shown live on the video feed

Setup (run one-time)

# 1. Create a virtual environment (recommended)
python -m venv venv
venv\Scripts\activate        # Windows

# 2. Install dependencies
pip install -r requirements.txt

# 3. Run the project
python main.py

Controls

Key	Action
S	Save annotated screenshot + append measurements to CSV log
P	Toggle birds-eye perspective correction (needs A4 paper flat in frame)
Q	Quit

Output files

output/
 - screenshots/
   - measurement_20250618_143022.png  
 -  logs/
    - measurements_2025-06-18.csv       

The timestamp in the screenshot filename matches the rows in the CSV so you can trace every measurement back to its source image.

Tips for best results

• Lighting: Even, diffused light (no harsh shadows should be made). Natural daylight is ideal.
• Background: A solid-colour surface gives the clearest edges.
• Card placement: Put the ID card on the left side of the frame laying flat.
• Perspective mode: Place an A4 sheet flat as a mat, put card + objects on top, press P.

Project structure

object_measurer/
- main.py                  <- entry point & main loop
- config.py                <- all tuneable constants
- requirements.txt
- measurer/
   - detector.py          <- contour detection + PPM maths
   - perspective.py       <- birds-eye warp transform
   - visualizer.py        <- bounding boxes, labels, HUD
- utils/
    - logger.py             <- CSV measurement logging
    - saver.py              <- screenshot saving

Tweaking detection

Edit config.py if the detection is too noisy or missing objects:

Variable	Effect
MIN_CONTOUR_AREA	Increase to ignore small noise blobs
CANNY_THRESH1/2	Lower values detect more (faint) edges
GAUSSIAN_BLUR	Larger kernel = more smoothing (try (9,9) on a textured surface)
REF_ASPECT_TOLERANCE	Increase if card isn't being detected at a steep angle

Author

Anwesha Singh
B.Tech (Computer Science Engineering)
Manipal University Jaipur