Hand Sign Recognition Using Mediapipe and Machine Learning

This project implements a hand sign recognition system using a webcam and a pre-trained machine learning model. The system uses Mediapipe for hand landmark detection and a RandomForestClassifier model to predict hand signs.

Features

Real-time hand landmark detection using Mediapipe.
Hand sign recognition with a pre-trained RandomForestClassifier model.
Displays hand landmarks on the webcam feed.
Supports multiple hand sign classes.

Requirements

To run the project, you'll need the following dependencies:

Python Libraries

OpenCV (cv2) for capturing and processing webcam images.
Mediapipe (mediapipe) for detecting hand landmarks.
Scikit-learn (sklearn) for using the trained RandomForestClassifier model.
Numpy (numpy) for processing data in arrays.
Pickle (pickle) for loading the pre-trained model.

Install Dependencies

You can install the required libraries using pip:

pip install opencv-python mediapipe scikit-learn numpy

## How to Use

### Clone the Repository

Clone this repository to your local machine:

```bash
git clone <repo_url>

Download the Trained Model

Ensure that you have the trained model stored as model.p in the root directory of the project.

Run the Hand Sign Recognition

Run the inference_classifier.py script to start the hand sign recognition:

python inference_classifier.py

Using the Webcam

The script will access your webcam to capture frames in real-time.
Hand landmarks will be detected and visualized on the webcam feed.
The system will predict and print the corresponding hand sign based on the hand detected.

Exit the Program

Press q on your keyboard to exit the webcam feed and close the program.

File Structure

/data/                     # Directory containing training data images
/model.p                   # Trained model using RandomForestClassifier
/inference_classifier.py    # Main Python script for real-time hand sign recognition
/README.md                 # This file

Model Details

The hand sign recognition model is trained using a RandomForestClassifier from the Scikit-learn library.
The training data consists of images of hands performing different signs, and MediaPipe is used to extract hand landmarks from these images.
The model expects 126 features as input (x, y, z coordinates for 21 hand landmarks on both hands). If only one hand is detected, the input is padded with zeros to match the expected 126 features.

Labels

The current hand sign recognition model supports the following labels:

A for the sign representing the letter "A".
B for the sign representing the letter "B".
L for the sign representing the letter "L".

These labels can be extended by training the model with additional hand sign data.

Troubleshooting

`ValueError: X has 63 features, but RandomForestClassifier is expecting 126 features`

This error occurs when only one hand is detected. The code pads the data with zeros to ensure that the input size is 126 features, corresponding to both hands.

Webcam Not Detected:

Make sure that your webcam is properly connected and accessible. You can check your camera device using OpenCV by running the following simple capture script:

import cv2
cap = cv2.VideoCapture(0)
while True:
    ret, frame = cap.read()
    cv2.imshow('frame', frame)
    if cv2.waitKey(1) & 0xFF == ord('q'):
        break
cap.release()
cv2.destroyAllWindows()

Future Improvements

Extend the model to support more hand signs.
Optimize the detection for faster real-time processing.
Add a graphical interface to display predictions on the webcam feed.
Improve hand detection accuracy using dynamic image modes (non-static).

License

This project is open-source and available under the MIT License.

it22323934/Hand-Sign-Recognition