International Research Journal of Engineering and Technology (IRJET)
e-ISSN: 2395-0056
Volume: 11 Issue: 05 | May 2024
p-ISSN: 2395-0072
www.irjet.net
Sign Language Translation System Based on CNN Model Gayatri kanwade1, Vansh Koli2, Prof.Vijaykumar Shep3, Saish Purankar4, Vaishnav Shende5 *1,2,3,4,5Department Of Mechanical Engineering MIT School Of Engineering,
MIT-ADT University Pune, Maharashtra, India. ---------------------------------------------------------------------***---------------------------------------------------------------------
Abstract - This paper presents a project focused on
While there is considerable ongoing research in computer vision, particularly fueled by advancements in deep learning, there has been limited exploration into gesture recognition in Sign Language. Our paper aims to establish a baseline for Sign Language gesture identification and develop a model to enhance communication for individuals with speech impairments.
developing a deep learning model for classifying the diverse hand gestures used in sign language fingerspelling. The classification algorithm is trained using MNIST image data, with testing conducted on a varied dataset comprising realtime static photos. Deep learning techniques from TensorFlow, Keras, and machine learning libraries such as sklearn are employed, with the model based on Convolutional Neural Networks (CNN). The CNN model is pretrained using the MNIST dataset, and data augmentation techniques are applied to enhance accuracy, resulting in 99.71% training accuracy and 100% testing accuracy.
2. RELATED WORK In the paper titled "The Application of Deep Learning in Computer Vision" by Q. Wu, Y. Liu, Q. Li, S. Jin, and F. Li [1], the authors provide an overview of deep learning concepts and discuss various commonly used algorithms in computer vision. They also examine the current research landscape in computer vision, with a focus on the prevalent applications of deep learning in the field.
Key Words: Sign Language Translation, Sign Language Recognition, Convolutional Neural Networks, Image Processing, Deep Learning
1. INTRODUCTION
In the paper titled "Generalizing the Hough Transform to Detect Arbitrary Shapes" by D. Ballard [2], the Hough transform is introduced as a method for detecting curves by exploiting the duality between points on a curve and the parameters defining that curve. Initially restricted to binary edge images, subsequent work generalized the approach to detect some analytic curves in grey-level images, such as lines, circles, and parabolas. This involved establishing a mapping between picture space and Hough transform space, enabling the detection of instances of specific shapes within an image.
Communication with individuals who have hearing loss presents significant challenges. Hand gestures are commonly used by speech and hearing-impaired individuals for communication, creating a language barrier between them and non-impaired individuals. It is essential to develop systems capable of identifying various gestures and conveying information to the general population. Understanding sign language gestures can be difficult for many individuals, and finding interpreters when needed can be challenging. To address this issue, a prospective solution is proposed to translate hand positions and gestures from sign language in real-time. This solution involves an opensource web application accessible on any device equipped with a camera to capture hand positions and motions.
The paper "Distinctive Image Features from Scale-Invariant Key points" by D. G. Lowe [3] presents an approach for extracting invariant features from images, facilitating accurate matching across different viewpoints of objects or scenes. The method also discusses utilizing these features for object recognition, starting with individual feature comparisons against a database of recognized objects using a rapid nearest-neighbor method.
We explored various machine learning and deep learning techniques, including Support Vector Machines (SVM), Logistic Regression, K-nearest neighbors (KNN), and Convolutional Neural Networks (CNN), for sign language detection. Our investigation revealed that CNN is the most effective technique for constructing a sign language recognition model. This model is trained on standard hand gestures used in the Sign Language system, which facilitates communication for individuals with speech impairments. However, due to the complexity and diversity of these gestures, many people find them challenging to understand, leading to communication barriers between individuals with and without speech impairments.
© 2024, IRJET
|
Impact Factor value: 8.226
In the paper titled "Hand Gesture Recognition Using Otsu's Method" by V. Bhavana, G. M. Surya Mouli, and G. V. Lakshmi Lokesh [4], the authors propose a method for accurate hand motion recognition using computers and Arduino devices. The system involves preprocessing, segmentation, feature extraction, pixel shifting, and classification of RGB images captured by a laptop camera. Otsu's segmentation technique is applied to segment the collected RGB images into grayscale images and segment them into distinct regions.
|
ISO 9001:2008 Certified Journal
|
Page 304