Vaishali Kulkarni

Dr. Vaishali Kulkarni

Narsee Monjee Institute of Management Studies, Mumbai, Mahārāshtra, International
orcid

Research Focus & Metrics

Academic Career Length

17Years (2010-Present)

(Calculated once Scilit publication data is synced: current year − year of first publication + 1)

Representative Publications

View All Publications
1 November 2015
Verified
Conference Paper
Closed access
PBPurva C. Badhe
Purva C. Badhe
VKVaishali Kulkarni
Vaishali Kulkarni
Sign Language is a natural language which deaf community uses for communication. Sign Language (SL) is a subset of gestures or signs made with fingers, hands, arms, eyes, and head, face etc. Each gesture in SL has a meaning assigned to it. Understanding SL is nothing but understanding the meaning of these gestures. There exists a problem in communication when a person who completely relies on this gestural SL for communication tries to converse with a person who does not understand the SL. Every country has its own developed SL. In India, this language is called as "Indian Sign Language (ISL). This paper aims to develop an algorithm that will translate the ISL into English. This paper has implemented a system named as "Indian Sign Language (ISL) Translator using Gesture recognition algorithm". The system translates gestures made in ISL into English. The gestures that have been translated include numbers, alphabets and few phrases. The algorithm first performs data acquisition, then the pre-processing of gestures is performed to track hand movement using a combinational algorithm, and recognition is done using template matching. The database used for implementation has been self-created and includes total 130,000 videos; out of which 72,000 videos were used to create the system database and remaining 58,000 videos have been tested for checking the performance of the system. The accuracy of this system is as high as 97.5%.

1 May 2019 in Heliyon
Verified
Journal Article
Open access
SPShashikant Patil
Shashikant Patil
VKVaishali Kulkarni
Vaishali Kulkarni
ABArchana Bhise
Archana Bhise
AI techniques have lifelong impact in biomedics and widely accepted outcomes. The sole objective of the study is to evaluate accurate detection of caries using feature extraction and classification of the dental images along with amalgamation Adaptive Dragonfly algorithm (DA) algorithm and Neural Network (NN) classifier. Here proposed caries detection model is designed for detecting the tooth cavities in an accurate manner. This methodology has two main phases; feature extraction and classification. In all total 120 images database is split into three sets, randomly for evaluating the performance. Further, this categorization of the test cases aids in ensuring the enhancement of the performance. In each of the test cases, there are 40 caries images the investigation in the performance of the proposed caries detection model was done in terms of accuracy, sensitivity, specificity, and precision, FPR, FNR, NPV, FDR, F1Score and MCC. Here MPCA with Nonlinear Programming and Adaptive DA, the proposed model is termed as MNP-ADA. The performance of the proposed MPCA-ADA model is evaluated by comparing it over the other existing feature extraction models. MPCA-ADA over the conventional classifier models like PCA-ADA, LDA-ADA and ICA-ADA in terms of performance parameters and MCC for all the test types and have superior results than the existing ones. The research work emphasizes the prospective efficacy of IP and NN algorithms for the detection and diagnosis of dental caries. A novel and improved model shows substantially worthy performance in distinguishing dental caries using image processing techniques.

13 November 2020 in International Journal of Speech Technology
Verified
Journal Article
Closed access
PDF file
SNSumita Nainan
Sumita Nainan
VKVaishali Kulkarni
Vaishali Kulkarni

Verified
Journal Article
Open access
PDF file
KBKavita Piyush Bani
Kavita Piyush Bani
VKVaishali Kulkarni
Vaishali Kulkarni
Day by day, the demand for wireless systems is increasing while the available spectrum resources are not sufficient. To fulfil the demand for wireless systems, the spectrum hole (spectrum vacant) should be found and utilised very effectively. Cognitive radio (CR) is a device which intelligently senses the spectrum through various spectrum-sensing detectors. Based on the complexity and licensed user’s information present with CR, the appropriate detector should be utilised for spectrum sensing. In this paper, a hybrid detector (HD) is proposed to determine the spectrum hole from the available spectrum resources. HD is designed based on an energy detector (ED) and matched detector (MD). Unlike a single detector such as ED or MD, HD can sense the signal more precisely. Here, HD can work on both conditions whether the primary user (PU) information is available or not. HD is analysed under heterogeneous environments with and without cooperative spectrum sensing (CSS). For CSS, four users were used to implement OR, AND, and majority schemes under low SNR walls. To design the HD, specifications were chosen based on the IEEE Wireless Regional Area Network (WRAN) 802.22 standard for accessing TV spectrum holes. For the HD model, we achieved the best results through OR rule. Under the low SNR circumstances at −20 dB SNR, the probability of detection (PD) is maximised to 1 and the probability of a false alarm (PFA) is reduced to 0 through the CSS environment.

1 March 2019
Verified
Conference Paper
Closed access
PSPranali Shenvi
Pranali Shenvi
NSNeel Samant
Neel Samant
SKShubham Kumar
Shubham Kumar
VKVaishali Kulkarni
Vaishali Kulkarni
Changing technology leads to various advancements. Here we talk about online transactions that is by the means of credit cards, which leads to credit card frauds, and this paper is about making a credit card fraud detection system using Deep Learning Neural Networks. Even if the Neural Network is trained over a large number of iterations, it is not accurate enough to classify the data as fraudulent or non-fraudulent because of the skewness in the dataset. We use two sampling mechanisms: Under-Sampling, by reducing number of non-fraudulent observations and Over-Sampling, where the fraudulent class observations are duplicated. The results obtained show that these methods increase the accuracy of the prediction process.

1 April 2014
Verified
Conference Paper
Closed access
PNPurva A. Nanivadekar
Purva A. Nanivadekar
VKVaishali Kulkarni
Vaishali Kulkarni
In this paper a Sign Language Recognition system has been proposed. The first step of this system is to create a database of Indian Sign Language. This is done by acquiring the videos from the signers while they are performing the hand gestures. Next step is Hand tracking and Segmentation. This is performed in order to extract features from a particular gesture. A three step algorithm has been used in the proposed system to get better quality hand tracking and segmentation. This algorithm works on motion tracking, edge detection and skin color detection. The system is implemented successfully and results are presented in this paper. The results demonstrate working of motion tracking, edge detection and skin color detection individually as well as their combined effect.