The World Wide Web Consortium describes web accessibility as the ability to access any information online regardless of the technology used and interacting with the interface of the web if required. Over the last few decades, significant efforts have been made in the field of web accessibility. Successful tools were developed leveraging technologies such as Artificial Intelligence, Speech Recognition systems and motion detection using Computer Vision. Sign Language is a form of non-verbal, hand movement or gesture based mode communicating thoughts and ideas. It is the primary way of communication for many people, majorly including differently abled people, such as people with speech-impairments. Despite being popular amongst people who are unable to speak or hear, there exists a great disparity in the knowledge of Sign Language in the general population consisting of able-bodied individuals. This necessitates the existence of a system that bridges this gap, and allows for efficient interpretation of Sign Language. Sign Language recognition is a growing field of interest in the domain of Artificial Intelligence, and many tools have been created to automate this process using image recognition techniques. These techniques are used to classify sign language and map them into their corresponding meaning in the English language. The steps required in this include the acquisition of data, extraction of relevant features followed by the training of the classifier. This paper gives an overview of the level of work done in this field, and delineates the scope of some models created for this purpose. It also discusses the challenges that are being faced and the scope of work that is yet to be done.

A review on Artificial Intelligence based Sign Language Recognition Techniques

The World Wide Web Consortium describes web accessibility as the ability to access any information online regardless of the technology used and interacting with the interface of the web if required. Over the last few decades, significant efforts have been made in the field of web accessibility. Successful tools were developed leveraging technologies such as Artificial Intelligence, Speech Recognition systems and motion detection using Computer Vision. Sign Language is a form of non-verbal, hand movement or gesture based mode communicating thoughts and ideas. It is the primary way of communication for many people, majorly including differently abled people, such as people with speech-impairments. Despite being popular amongst people who are unable to speak or hear, there exists a great disparity in the knowledge of Sign Language in the general population consisting of able-bodied individuals. This necessitates the existence of a system that bridges this gap, and allows for efficient interpretation of Sign Language. Sign Language recognition is a growing field of interest in the domain of Artificial Intelligence, and many tools have been created to automate this process using image recognition techniques. These techniques are used to classify sign language and map them into their corresponding meaning in the English language. The steps required in this include the acquisition of data, extraction of relevant features followed by the training of the classifier. This paper gives an overview of the level of work done in this field, and delineates the scope of some models created for this purpose. It also discusses the challenges that are being faced and the scope of work that is yet to be done. 

Indian Sign Language is the primary mode of communication known to persons who use Indian Sign Language for people who have hearing or language deficits. Different types of machine learning models are used to broaden the scope of communication for those with impairments and illiteracy. There are numerous machine learning models for analyzing gestures, postures, and facial recognition in Indian Sign Language for single-handed and double-handed signals. The present study on hand gestures, recognition, and translation intends to build an essential foundation for developing a platform to facilitate communication for the s pecially-abled with anyone. Machine learning algorithms generally focus on letter recognition or a few fundamental indicators. Communication is essential for exchanging ideas, thoughts, and feelings. Sign language is a kind of communication that uses hand motions. This is aimed toward those with impairments such as muteness and deafness. Machine learning, a branch of artificial intelligence, will aid in identifying various hand motions and predicting the language created by those inputs based on those inputs[2]. Sign language has a grammar that is unique from and independent of English. When compared to English, SL allows for far more freedom in word order. Tense is marked morphologically on verbs in English, but SL (like many other languages, such as Indian Sign Language) communicates tense lexically using temporal adverbs. The structure of ISL and English differs at the phonological level as well. Signed languages, like spoken languages, include a degree of sublexical structure that includes segments and combinatorial rules; however, phonological elements are manual rather than vocal. The way spatial information is conveyed in English and ISL differs substantially. All Deaf people are illiterate in written English. As an output, the SL text can be produced. SL is just physically executed English, where English and SL share the identical linguistic structure-that one is a straight encoding of the other. Many software designers mistakenly believe that deaf users can always access printed the English language in a user interface. Many designers feel that if auditory information is also supplied as written English, the deaf user's demands will be addressed. Prepositions such as “in,” “on,” and “under” are used to indicate locative information in English, as in many other spoken languages. On the other hand, SL encodes locative and motion information via verbal classifier formulations in which hand shape morphemes define item type, and the location of the hands in signing space schematically depicts the spatial relationship between two things. Thus, English and ASL differ significantly in phonological, morphological, and syntactic areas.

A Systematic Review of Machine Learning Approaches for Classifying Indian Sign Language Gestures and Facial Expressions

Kinesics has a decisive role for the mute and deaf for inter personal communication. It necessitates body postures, face expressions and movement of hands Due to the diversity of population, each region has genus of sign language based on its individual rule of grammar. The objective is to manufacture a dynamic gesture of hand identification artifice for Indian kinesics using morphological algorithms. The pithy frames are extricated from the captured video for feature extrication and the interpretation of the sign is displayed on the screen in text format.

Morphological Based Dynamic Hand Gesture Recognition for Indian Sign Language

Nowadays, dynamic hand gesture recognition has become a complicated work in the recognition of pattern and the communities with consideration of computer vision. This paper tempts to frame an algorithmic analysis on proposed “static and dynamic-oriented hand gesture recognition”. Moreover, the recognition of static and dynamic models is improved by Dynamic Time Warping (DTW) model. In the phase called pre-processing, “grey scale conversion and histogram equalization” is used, whereas, in segmentation, “Active Contour model, and canny edge detection” is employed. Moreover, the significant features are extracted depending on the “Histogram of Oriented Gradients (HOG), and Edge Oriented Histogram (EOH)”, and the feature dimension is reduced by Principle Component Analysis (PCA). The optimal feature selection technique is adopted by the novel “Deer Hunting-based Grey Wolf Optimization (DH-GWO)”. Moreover, the “significant frames in the video” are eliminated by the DTW pattern. Finally, the characters and words are exactly recognized by Neural Network (NN), using the DH-GWO training. Here, the analysis is carried out by varying the random number $e$ from 0.5 to 3.0 of the proposed DH-GWO.

Algorithmic Analysis on Deer Hunting-based Grey Wolf for Dynamic Time Wrapping-based Hand Gesture Recognition

Enhanced Nature-Inspired Algorithm-based Hybrid Deep Learning for Character Recognition in Sanskrit Language

Accidents and road crashes occurring due to dissociation of driver from his consciousness is one of the hot topics researchers have been contemplating to curtail, from the past few years. This dissociation wanes the driver's ability to concentrate on the road while driving and promotes occurrence of mishaps. There has been a challenge to non-intrusively detect drowsiness and fatigue-level of drivers in real-time environments and concomitantly make it computationally faster and less complex. The paper proposes a faster real-time driver inactiveness detection system capable of alerting drivers of their drowsiness by inspecting and scrutinizing the state of eyes and mouth and follows a non-obtrusive approach. The system utilizes eyes to monitor the drooping nature of eyes and examines the mouth's current state to detect any yawning activity performed. Initially, it uses Haar transform to extract facial features and strong classifiers for detecting faces by incorporating Adaboost algorithm. Next, shape predictors are used to locate the key points for obtaining position of mouth and eyes within the face area. States of the mouth and eyes are continuously monitored with the help of the metrics viz. Eye Aspect Ratio (EAR) and Mouth Yawn Width respectively and are used to alert the driver through voice-enabled commands followed by a warning alarm sound to trigger them thereby aiding in regaining their consciousness back and focusing on their driving. The algorithm used in this system is tested under different lighting conditions and against different possible situations and accurate results in terms of successful detection rates are obtained.

Manisha Ravikumar Kowdiki

Papers

Morphological Based Dynamic Hand Gesture Recognition for Indian Sign Language

Algorithmic Analysis on Deer Hunting-based Grey Wolf for Dynamic Time Wrapping-based Hand Gesture Recognition

Enhanced Nature-Inspired Algorithm-based Hybrid Deep Learning for Character Recognition in Sanskrit Language

Real-Time Voice Enabled Drivers' Safety Monitoring System Using Image Processing

An Attention-Based Swin U-Net-Based Segmentation and Hybrid Deep Learning Based Diabetic Retinopathy Classification Framework Using Fundus Images