scispace - formally typeset
Search or ask a question
Author

Annapurna Sharma

Other affiliations: Dongseo University
Bio: Annapurna Sharma is an academic researcher from International Institute of Information Technology, Bangalore. The author has contributed to research in topics: Handwriting recognition & Handwriting. The author has an hindex of 6, co-authored 12 publications receiving 131 citations. Previous affiliations of Annapurna Sharma include Dongseo University.

Papers
More filters
Proceedings ArticleDOI
11 Nov 2008
TL;DR: A 4-layer back propagation neural network, with Levenberg-marquardt algorithm for training, showed best performance among the other neural network training algorithms.
Abstract: This paper presents the designing of a neural network for the classification of Human activity. A Tri-axial accelerometer sensor, housed in a chest worn sensor unit, has been used for capturing the acceleration of the movements associated. All the three axis acceleration data were collected at a base station PC via a CC2420 2.4 GHz ISM band radio (zigbee wireless compliant), processed and classified using MATLAB. A neural network approach for classification was used with an eye on theoretical and empirical facts. The work shows a detailed description of the designing steps for the classification of human body acceleration data. A 4-layer back propagation neural network, with Levenberg-marquardt algorithm for training, showed best performance among the other neural network training algorithms.

44 citations

Proceedings ArticleDOI
14 Oct 2008
TL;DR: In this paper, Rest, walking and running are the classified activities of the person and both time and frequency analysis was performed to classify running and walking.
Abstract: Activity classification was performed using MEMS accelerometer and wireless sensor node for wireless sensor network environment. Three axes MEMS accelerometer measures body's acceleration and transmits measured data with the help of sensor node to base station attached to PC. On the PC, real time accelerometer data is processed for movement classifications. In this paper, Rest, walking and running are the classified activities of the person. Both time and frequency analysis was performed to classify running and walking. The classification of rest and movement is done using Signal magnitude area (SMA). The classification accuracy for rest and movement is 100%. For the classification of walk and Run two parameters i.e. SMA and Median frequency were used. The classification accuracy for walk and running was detected as 81.25% in the experiments performed by the test persons.

41 citations

Proceedings ArticleDOI
10 Oct 2008
TL;DR: In this article, the frequency of the body acceleration data of the three axes for classifying the activities in a set of data is used. But the algorithm includes a normalization step and hence there is no need to set a different value of threshold value for magnitude for different test person.
Abstract: This work presents, the classification of user activities such as Rest, Walk and Run, on the basis of frequency component present in the acceleration data in a wireless sensor network environment. As the frequencies of the above mentioned activities differ slightly for different person, so it gives a more accurate result. The algorithm uses just one parameter i.e. the frequency of the body acceleration data of the three axes for classifying the activities in a set of data. The algorithm includes a normalization step and hence there is no need to set a different value of threshold value for magnitude for different test person. The classification is automatic and done on a block by block basis.

26 citations

Journal ArticleDOI
TL;DR: A fully convolution based deep network architecture for cursive handwriting recognition from line level images that has fewer parameters and takes less training and testing time, making it suitable for low-resource and environment-friendly deployment.
Abstract: Recognition of cursive handwritten images has advanced well with recent recurrent architectures and attention mechanism. Most of the works focus on improving transcription performance in terms of Character Error Rate (CER) and Word Error Rate (WER). Existing models are too slow to train and test networks. Furthermore, recent studies have recommended models be not only efficient in terms of task performance but also environmentally friendly in terms of model carbon footprint. Reviewing the recent state-of-the-art models, it recommends considering model training and retraining time while designing. High training time increases costs not only in terms of resources but also in carbon footprint. This becomes challenging for handwriting recognition model with popular recurrent architectures. It is truly critical since line images usually have a very long width resulting in a longer sequence to decode. In this work, we present a fully convolution based deep network architecture for cursive handwriting recognition from line level images. The architecture is a combination of 2-D convolutions and 1-D dilated non causal convolutions with Connectionist Temporal Classification (CTC) output layer. This offers a high parallelism with a smaller number of parameters. We further demonstrate experiments with various re-scaling factors of the images and how it affects the performance of the proposed model. A data augmentation pipeline is further analyzed while model training. The experiments show our model, has comparable performance on CER and WER measures with recurrent architectures. A comparison is done with state-of-the-art models with different architectures based on Recurrent Neural Networks (RNN) and its variants. The analysis shows training performance and network details of three different dataset of English and French handwriting. This shows our model has fewer parameters and takes less training and testing time, making it suitable for low-resource and environment-friendly deployment.

25 citations

Proceedings ArticleDOI
01 Aug 2018
TL;DR: A system capable of taking the input as handwritten essays in image format and outputs the grading on the scale of 0-5; 0 being the worst and 5 being the best, which indicates that the current OHR systems have transcription errors but as a whole can perform well for an application like AES.
Abstract: Automatic grading of handwritten essays is vital in evaluating the performance of students in educational settings, particularly in situations where language experts are rare. We build a system capable of taking the input as handwritten essays in image format and outputs the grading on the scale of 0-5; 0 being the worst and 5 being the best. The overall system integrates Optical Handwriting Recognition (OHR) and Automated Essay Scoring (AES)/grading. The handwritten essay is transcribed using a network composed of Multi-Dimensional Long Short Term Memory (MDLSTM) and convolution layers. The loss function is Connectionist Temporal Classification (CTC). The AES model is a 2-layer artificial neural network with a feature set based on pretrained GloVe word vectors. The results of grading of essays are compared for transcriptions of essays received from OHR system and transcriptions of essays done manually. The mutual agreement between the two shows a Quadratic Weighted Kappa score of 0.88. The results indicate that though the current OHR systems have transcription errors but as a whole can perform well for an application like AES.

15 citations


Cited by
More filters
Journal ArticleDOI
TL;DR: A deep convolutional neural network is proposed to perform efficient and effective HAR using smartphone sensors by exploiting the inherent characteristics of activities and 1D time-series signals, at the same time providing a way to automatically and data-adaptively extract robust features from raw data.
Abstract: This paper proposes a deep convolutional neural network for HAR using smartphone sensors.Experiments show that the proposed method derives relevant and more complex features.The method achieved an almost perfect classification on moving activities.It outperforms other state-of-the-art data mining techniques in HAR. Human activities are inherently translation invariant and hierarchical. Human activity recognition (HAR), a field that has garnered a lot of attention in recent years due to its high demand in various application domains, makes use of time-series sensor data to infer activities. In this paper, a deep convolutional neural network (convnet) is proposed to perform efficient and effective HAR using smartphone sensors by exploiting the inherent characteristics of activities and 1D time-series signals, at the same time providing a way to automatically and data-adaptively extract robust features from raw data. Experiments show that convnets indeed derive relevant and more complex features with every additional layer, although difference of feature complexity level decreases with every additional layer. A wider time span of temporal local correlation can be exploited (1?9-1?14) and a low pooling size (1?2-1?3) is shown to be beneficial. Convnets also achieved an almost perfect classification on moving activities, especially very similar ones which were previously perceived to be very difficult to classify. Lastly, convnets outperform other state-of-the-art data mining techniques in HAR for the benchmark dataset collected from 30 volunteer subjects, achieving an overall performance of 94.79% on the test set with raw sensor data, and 95.75% with additional information of temporal fast Fourier transform of the HAR data set.

854 citations

Journal ArticleDOI
TL;DR: This is one of the first surveys to provide such breadth of coverage across different wearable sensor systems for activity classification, and found that these single sensing modalities laid the foundation for hybrid works that tackle a mix of global and local interaction-type activities.
Abstract: Activity detection and classification are very important for autonomous monitoring of humans for applications, including assistive living, rehabilitation, and surveillance. Wearable sensors have found wide-spread use in recent years due to their ever-decreasing cost, ease of deployment and use, and ability to provide continuous monitoring as opposed to sensors installed at fixed locations. Since many smart phones are now equipped with a variety of sensors, such as accelerometer, gyroscope, and camera, it has become more feasible to develop activity monitoring algorithms employing one or more of these sensors with increased accessibility. We provide a complete and comprehensive survey on activity classification with wearable sensors, covering a variety of sensing modalities, including accelerometer, gyroscope, pressure sensors, and camera- and depth-based systems. We discuss differences in activity types tackled by this breadth of sensing modalities. For example, accelerometer, gyroscope, and magnetometer systems have a history of addressing whole body motion or global type activities, whereas camera systems provide the context necessary to classify local interactions, or interactions of individuals with objects. We also found that these single sensing modalities laid the foundation for hybrid works that tackle a mix of global and local interaction-type activities. In addition to the type of sensors and type of activities classified, we provide details on each wearable system that include on-body sensor location, employed learning approach, and extent of experimental setup. We further discuss where the processing is performed, i.e., local versus remote processing, for different systems. This is one of the first surveys to provide such breadth of coverage across different wearable sensor systems for activity classification.

320 citations

Journal ArticleDOI
TL;DR: This paper adopts Random Forest to select the important feature in classification and compares the result of the dataset with and without essential features selection by RF methods varImp(), Boruta, and Recursive Feature Elimination to get the best percentage accuracy and kappa.
Abstract: Feature selection becomes prominent, especially in the data sets with many variables and features. It will eliminate unimportant variables and improve the accuracy as well as the performance of classification. Random Forest has emerged as a quite useful algorithm that can handle the feature selection issue even with a higher number of variables. In this paper, we use three popular datasets with a higher number of variables (Bank Marketing, Car Evaluation Database, Human Activity Recognition Using Smartphones) to conduct the experiment. There are four main reasons why feature selection is essential. First, to simplify the model by reducing the number of parameters, next to decrease the training time, to reduce overfilling by enhancing generalization, and to avoid the curse of dimensionality. Besides, we evaluate and compare each accuracy and performance of the classification model, such as Random Forest (RF), Support Vector Machines (SVM), K-Nearest Neighbors (KNN), and Linear Discriminant Analysis (LDA). The highest accuracy of the model is the best classifier. Practically, this paper adopts Random Forest to select the important feature in classification. Our experiments clearly show the comparative study of the RF algorithm from different perspectives. Furthermore, we compare the result of the dataset with and without essential features selection by RF methods varImp(), Boruta, and Recursive Feature Elimination (RFE) to get the best percentage accuracy and kappa. Experimental results demonstrate that Random Forest achieves a better performance in all experiment groups.

271 citations

Journal ArticleDOI
TL;DR: The reliability and validity outcomes suggest that a number of measurement systems and testing procedures can be implemented to accurately assess maximum strength and ballistic performance in recreational and elite athletes, alike, but the reader needs to be cognisant of the inherent differences between measurement systems.
Abstract: An athletic profile should encompass the physiological, biomechanical, anthropometric and performance measures pertinent to the athlete’s sport and discipline. The measurement systems and procedures used to create these profiles are constantly evolving and becoming more precise and practical. This is a review of strength and ballistic assessment methodologies used in sport, a critique of current maximum strength [one-repetition maximum (1RM) and isometric strength] and ballistic performance (bench throw and jump capabilities) assessments for the purpose of informing practitioners and evolving current assessment methodologies. The reliability of the various maximum strength and ballistic assessment methodologies were reported in the form of intra-class correlation coefficients (ICC) and coefficient of variation (%CV). Mean percent differences $$ \left( {M_{\text{diff}} = \left[ {\frac{{{\mid }X_{{{\text{method}}1}} - X_{{{\text{method}}2}} {\mid }}}{{(X_{{{\text{method}}1}} + X_{{{\text{method}}2}} )}}} \right] \times 100} \right) $$ and effect size (ES = [X method2 − X method1] ÷ SDmethod1) calculations were used to assess the magnitude and spread of methodological differences for a given performance measure of the included studies. Studies were grouped and compared according to their respective performance measure and movement pattern. The various measurement systems (e.g. force plates, position transducers, accelerometers, jump mats, optical motion sensors and jump-and-reach apparatuses) and assessment procedures (i.e. warm-up strategies, loading schemes and rest periods) currently used to assess maximum isometric squat and mid-thigh pull strength (ICC > 0.95; CV 0.91; CV 0.82; CV < 6.5 %) were deemed highly reliable. The measurement systems and assessment procedures employed to assess maximum isometric strength [M Diff = 2–71 %; effect size (ES) = 0.13–4.37], 1RM strength (M Diff = 1–58 %; ES = 0.01–5.43), vertical jump capabilities (M Diff = 2–57 %; ES = 0.02–4.67) and bench throw capabilities (M Diff = 7–27 %; ES = 0.49–2.77) varied greatly, producing trivial to very large effects on these respective measures. Recreational to highly trained athletes produced maximum isometric squat and mid-thigh pull forces of 1,000–4,000 N; and 1RM bench press, back squat and power clean values of 80–180 kg, 100–260 kg and 70–140 kg, respectively. Mean and peak power production across the various loads (body mass to 60 % 1RM) were between 300 and 1,500 W during the bench throw and between 1,500 and 9,000 W during the vertical jump. The large variations in maximum strength and power can be attributed to the wide range in physical characteristics between different sports and athletic disciplines, training and chronological age as well as the different measurement systems of the included studies. The reliability and validity outcomes suggest that a number of measurement systems and testing procedures can be implemented to accurately assess maximum strength and ballistic performance in recreational and elite athletes, alike. However, the reader needs to be cognisant of the inherent differences between measurement systems, as selection will inevitably affect the outcome measure. The strength and conditioning practitioner should also carefully consider the benefits and limitations of the different measurement systems, testing apparatuses, attachment sites, movement patterns (e.g. direction of movement, contraction type, depth), loading parameters (e.g. no load, single load, absolute load, relative load, incremental loading), warm-up strategies, inter-trial rest periods, dependent variables of interest (i.e. mean, peak and rate dependent variables) and data collection and processing techniques (i.e. sampling frequency, filtering and smoothing options).

204 citations

Journal ArticleDOI
07 Mar 2017-Sensors
TL;DR: This paper creates the most complete dataset, focusing on accelerometer sensors, and conducts an extensive analysis on feature representations and classification techniques (the most comprehensive comparison yet with 293 classifiers) for activity recognition.
Abstract: Sensor-based motion recognition integrates the emerging area of wearable sensors with novel machine learning techniques to make sense of low-level sensor data and provide rich contextual information in a real-life application. Although Human Activity Recognition (HAR) problem has been drawing the attention of researchers, it is still a subject of much debate due to the diverse nature of human activities and their tracking methods. Finding the best predictive model in this problem while considering different sources of heterogeneities can be very difficult to analyze theoretically, which stresses the need of an experimental study. Therefore, in this paper, we first create the most complete dataset, focusing on accelerometer sensors, with various sources of heterogeneities. We then conduct an extensive analysis on feature representations and classification techniques (the most comprehensive comparison yet with 293 classifiers) for activity recognition. Principal component analysis is applied to reduce the feature vector dimension while keeping essential information. The average classification accuracy of eight sensor positions is reported to be 96.44% ± 1.62% with 10-fold evaluation, whereas accuracy of 79.92% ± 9.68% is reached in the subject-independent evaluation. This study presents significant evidence that we can build predictive models for HAR problem under more realistic conditions, and still achieve highly accurate results.

162 citations