Convolutional Neural Network based Bidirectional Sign Language Translation System

Fernandes, Lance; Dalvi, Prathamesh; Junnarkar, Akash; Bansode, Manisha

doi:10.1109/icssit48917.2020.9214272

Cited by 19 publications

(2 citation statements)

References 8 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…We have discussed both approaches: vision based and sensor/glove-based. Glove-based or Sensor-based approaches are used by [1][2][3][4][5]. In this approach, users may be required to wear gloves/sensors (wire or wireless).…”

Section: Related Workmentioning

confidence: 99%

See 1 more Smart Citation

Use of Key Points and Transfer Learning Techniques in Recognition of Handedness Indian Sign Language

Ingoley,

Bakal

2023

IJRITCC

View full text Add to dashboard Cite

The most expressive way of communication for individuals who have trouble speaking or hearing is sign language. Normal people are unable to comprehend sign language. As a result, communication barriers are put up. Majority of people are right-handed. Statistics say that, an average population of left-handed person in the world is about 10%, where they use left hand as their dominating hand. In case of hand written text recognition, if the text is written by left-handed or right-handed person, then there would not be any problem in recognition neither for human and nor for computer. But same thing is not true for sign language and its detection using computer. When the detection is performed using computer vision and if it falls into the category of detection by appearance, then it might not detect correctly. In machine and deep learning, if the model is trained using just one dominating hand, let’s say right hand, then the predictions can go wrong if same sign is performed by left-handed person. This paper addresses this issue. It takes into account the signs performed by any type of signer: left-handed, right-handed or ambidexter. In proposed work is on Indian Sign Language (ISL). Two models are trained: Model I, is trained on one dominating hand and Model II, is trained on both the hands. Model II gives correct predictions regardless of any type of signer. It recognizes alphabets and numbers in ISL. We used the concept of Key points and Transfer Learning techniques for implementation. Using this approach, models get trained quickly and we could achieve validation accuracy of 99%.

show abstract

Section: Related Workmentioning

confidence: 99%

“…are part of their glove-based system. To improve the efficiency of hardware glove-based system [2] uses CNN and also created self-dataset. They developed bidirectional sign language translating system i) sign to text and text to speech and ii) speech to text and text to sign.…”

Section: Related Workmentioning

confidence: 99%