Goal is to recognize hand gestures. I've trained the model on my own dataset using Perceptron. I've included my the dataset in the repository itself. In it's present state the model is trained to recognize just five gestures but can easily be trained for many hand gestures. I'll also upload the code that I'm using for capturing the hand an processing it for training the model.
Model gives a high testing accuracy of about 95% using just two convolution and three Dense layers. But if you want to train more hand gestures then you'll probably need more a complex network.
Images in the dataset are of dimension 275 by 275. But for performance reasons they have been resized to 50 by 50. You can use them as it is if you have a powerful setup otherwise program displays a Tensorflow error tensorflow/core/framework/allocator.cc:101] Allocation of X exceeds 10% of system memory.
capturing hand images.ipynb- This program can capture new hand gestures and write them in the specified directory.combining images.ipynb- This program combines the images in a single array.recognizer.ipynb- This is the main program that uses pretrained model (in the repo) for recognizing hand gestures.demo.mp4- This is the video that shows the working of the model in real time.sample images- This folder contains the samle images used for training the model.loading dataset and training model.ipynb- This program uses the given dataset to train the CNN model.hand dataset.rar- This rar file have the dataset used to train and test the CNN model.loading dataset and training model.ipynb- This program uses the given dataset to train the CNN model.
cv2 (opencv)pickleglobsklearn (scikit-learn)kerasnumpy
I hope to implement more than five gestures in the future. There will be further improvements in the code itself too.
As of Jan, 2020 I have used CNN for the recognition of 5 gestures.
I've used both the Perceptron as well as the CNN Model for recognition of 5 hand gestures. And I can easily say CNN works better for extracting features of an image. There's and improvement of Error %age from 32% to 5%. This happens because Perceptron model is a simple 1 dimensional set of neurons, therefore it reduces a lot of features in the images whereas the CNN is specially designed to work with images and works on 2 dimensional set of neurons.