Build a Real-Time Indian Sign Language (ISL) Recognition Model

Job ID: 40054066

Budget: ₹1,500 – ₹12,500 INR

Project Title:

Build a Real-Time Indian Sign Language (ISL) Recognition Model in 1–2 Days (Words, Characters, Numbers)

Project Description:

I need a machine learning engineer to quickly develop a trained model capable of detecting Indian Sign Language (ISL) gestures, specifically words, characters (A-Z), and numbers (0-9), from live video input. The model should be capable of recognizing these gestures in real-time (within 100ms latency).

Key Features:

Character Detection (A–Z):
Recognize hand gestures representing the 26 English alphabets in Indian Sign Language (ISL).

Number Recognition (0–9):
Recognize numbers represented in ISL (0–9).

Word Recognition:
Recognize common Indian Sign Language (ISL) words like hello, thank you, sorry, please, yes, no, love, you, me, and others.

Mixed Gesture Sequences:
The model should recognize and output mixed gestures like:

"I love you 3000" → "I" (character) + "love" (word) + "you" (word) + "3000" (number).

Real-Time Detection:
The model should be optimized for real-time use (e.g., < 100ms detection latency) for live webcam input.

Deliverables:

Trained Model:

A trained machine learning model capable of detecting alphabets, numbers, and common words in ISL.

Model export formats: .h5, .onnx, .tfjs (for web).

Dataset:

If the dataset is not already available, provide guidance on how to quickly collect and label a small dataset of hand gestures (e.g., A-Z, 0-9, and common words).

At least 100-200 samples per gesture would be sufficient to start.

Preprocessing Code:

Code to preprocess video frames, including hand detection and feature extraction (e.g., hand landmarks, image normalization).

Real-Time Inference Code:

Python or TensorFlow.js code that processes webcam video in real-time and uses the trained model for gesture recognition.

Code should include token detection, sentence building, and displaying the recognized text.

Documentation:

A basic README explaining the model architecture, how to run the inference code, and usage instructions.

Required Skills:

Machine Learning (ML) with experience in computer vision and gesture recognition.

Deep Learning frameworks such as TensorFlow, Keras, or PyTorch.

Familiarity with real-time gesture recognition models (e.g., MediaPipe, HandPose, OpenCV).

Experience with TensorFlow.js for web-based deployment is a plus.

Ability to work under tight deadlines and deliver working code quickly.

Preferred Experience:

Previous work on gesture recognition, sign language, or pose estimation models.

Knowledge of real-time video processing and model optimization for speed.

Timeline:

1-2 Days: This is a time-sensitive project. The goal is to have a prototype or MVP delivered within 1-2 days.

Quick Turnaround: Please provide an estimate based on this aggressive timeline.

Budget:

The budget is negotiable based on your expertise and quick turnaround. Please provide an estimated cost for the entire project along with a detailed breakdown of tasks.

How to Apply:

Please include:

A brief summary of your relevant experience in gesture recognition and real-time models.

Your proposed approach to building the model and meeting the deadline.

An estimated cost and timeline (with consideration for the 1-2 day requirement).

Examples of similar projects or portfolios (if available).

Additional Notes:

If you have a pre-existing dataset or can quickly adapt an existing one for ISL, please mention it.

The project will require real-time inference, so please optimize for low latency (less than 100 ms).

Looking forward to your proposals!