Sign Language: Train an ASL gesture recognition CNN

Understand the project

Practical overview

Use it as a code-first starting point when learning or prototyping ASL gesture image classification and willing to work directly with scripts.

Learners exploring CNNs and image classification through an ASL gesture project.
Developers prototyping gesture-image capture, dataset preparation and model training.

Capture and augment image samples for gesture categories.
Train a CNN using TensorFlow or Keras.
Explore recognition of trained gestures from a video stream using the Keras-trained model.
Generate classification reports, including a confusion matrix, F-scores, precision and recall.

Before you choose

The README lists Python 3.x, TensorFlow 1.5, Keras, OpenCV 3.4, h5py and pyttsx3.
CPU and NVIDIA GPU dependency files are provided; a capable CPU is recommended, preferably with a GPU.
The README describes 44 existing gesture categories, with 1,200 grayscale 50×50 captures per gesture and flipped images for augmentation.
The README describes video-stream recognition with the Keras-trained model, but reports difficulty using the TensorFlow-trained model for prediction.
The repository has no interactive interface, and expansion may require script and model changes.

Check current requirements in the official documentation. Source-based guidance does not mean OpenViST has run or independently tested this project.

Source-based practical guide

Capture images for a new or replacement gesture and prepare them for model training.

Python 3.x and the project’s listed TensorFlow 1.5, Keras, OpenCV 3.4, h5py and pyttsx3 dependencies.
A webcam and familiarity with the project scripts.
A capable CPU; the README prefers a GPU and provides separate CPU and NVIDIA GPU requirements files.

Run the command below and follow the window prompts to set and save the hand histogram; repeat if lighting conditions change.
Documented command
```
python set_hand_hist.py
```
Run the command below, enter a gesture number and name, and capture images in the indicated webcam window.
Documented command
```
python create_gestures.py
```
Run the command below to flip the captured images, then run the command below to load the images.
Documented command
```
python flip_images.py
```
Documented command
```
python load_images.py
```

A captured and loaded gesture dataset ready for the project’s model-training stage.

The README warns that the repository has no interactive interface and may require users to work out details or modify scripts.
The guide describes 1,200 captures per gesture and says capture closes automatically when the counter reaches 1,200.
Adding or removing gestures requires retraining; adding gesture classes may also require changes to network layers and parameters.

Evidence and transparency

Source record details

Source availability: Active
Sources checked: 18 October 2026
Original website title: EvilPort2/Sign-Language · Official homepage title · retained in its source language