Vaani, a repository dedicated to a significant project, serves as a platform for predicting and converting nonverbal sign language into text.
Engineered an innovative real-time sign language recognition system, integrating Machine Learning and advanced Deep Learning models. The ensemble includes Support Vector Machines (SVMs), Convolutional Neural Networks (CNNs), Long Short-Term Memory (LSTM), and Random Forest for robust analysis. Precise hand keypoint detection is achieved using Mediapipe, with enhanced data collection and naming features. The system, implemented in Python, leverages OpenCV and Mediapipe for efficient computer vision tasks. Trained on custom datasets, it accurately recognizes sign language gestures, now including word and sentence formation, with rapid detection of Hindi words. Optimal performance is ensured through extensive optimization efforts, highlighting a comprehensive understanding of sign language processing. The optimization efforts were directed towards achieving unparalleled real-time performance, underscoring a comprehensive understanding
Our objective is to attain the highest achievable accuracy for optimal results. We are actively engaged in ongoing enhancements within the project, continually refining the models based on insights gained from each iteration.
- CNN algorithm with specific no hand recognition technique:
- SVM algorithm used with canny edge detection, gray scaling and SIFT/SURF desc. algorithm:
- Using CVZone, mediapipe for better hand gesture recognition:
- Using CVZ along with training of array df(npy) using LSTM, Tensorflow and Keras:
- Words and Sentences formation with LSTM model:
https://github.com/praball/MAJOR_VAANI/assets/92200181/8dc44a7c-a8d6-495a-8bfe-3ff2a6a239ea
- Random Forest Implementation and rapid gesture conversion to Hindi words to form sentences:
