Intelligent Human Computer Interaction

Navigation of drones can be conceivably performed by operators by analyzing the brain signals of the person. EEG signal corresponding to the motor imaginations can be used for generation of control signals for drone. Different machine learning and deep learning approaches have been developed in the state of the art literature for the classification of motor imagery EEG signal. There is still a need for developing a suitable model that can classify the motor imagery signal fast and can generate a navigation command for drone in real-time. In this paper, we have reported the performance of convolutional stacked autoencoder and Convolutional Long short term memory models for classification of Motor imagery EEG signal. The developed models have been optimized using TensorRT that speeds up inference performance and the inference engine has been deployed on Jetson TX2 embedded platform. The performance of these models have been compared with different machine learning models.

EEG has proved to be a vital tool in diagnosing stroke. Now a days EEG is also used for tracking the rehabilitation process after stroke. Post stroke subjects are guided to do some physical activities to help them in regaining lost motor activity. The study presented here acquires EEG data from patients undergoing rehabilitation process and tracks the improvement in their motor ability. This is done by correlating change in Fugl-Meyer Assessment (FMA) score with some of the features obtained from EEG data. A significant correlation was found between FMA change and mean absolute value (r = 0.6, p < 0.001) of the EEG signal and also between change in mean alpha power ratio (left/right) vs. \(\varDelta \)FMA (r = −0.71, p < 0.001). The high correlation values of these features suggest that they can be used for monitoring rehabilitation after stroke.

This paper presents the development and evaluation of an algorithm for compressing fetal electrocardiographic signals, taken superficially on the mother’s abdomen. This method for acquiring ECG signals produces a great volumen of information that makes it difficult for the records to be stored and transmitted. The proposed algorithm aims for lossless compression of the signal by applying Wavelet Packet Transform to keep errors below the unit, with compression rates over 20:1 and with conserved energy in reconstruction as comparison parameter. For algorithm validation, the signal files provided by PhysioBank DataBase are used.

This paper aims to develop a Fuzzy Inference System that categorizes the Electroencephalographic (EEG) signals generated from a healthy brain with those generated by the brain suffering from epilepsy into different identifiable classes. This is done by statistical analysis of dynamical properties of the EEG signals using well established techniques for nonlinear time series analysis. For this purpose, a defined null hypothesis and obtained rejection counts are taken into consideration based on Randomness and Stationarity tests applied on the available EEG data. The outcome of this analysis is used to create a fuzzy inference model that can differentiate between brain states and zones they belong to viz. healthy zone, epileptic zone and non-epileptic zone. These zones are further classified into different states as eyes opened and closed for the healthy zone, the state where first ictal occurs, the seizure free interval and during the seizure activity within the epileptic zone and, the non focal and non epileptic state within the non–epileptic zone. Encouraging results have been obtained from this pilot project. However, limited data has been used for the development work and hence the decisions may be specific. Further generalization of the rules is possible with the help of more data and inputs from neurophysiological experts.

Brain state classification using electroencephalography (EEG) finds applications in both clinical and non-clinical contexts, such as detecting sleep states or perceiving illusory effects during multisensory McGurk paradigm, respectively. Existing literature mostly considers recordings of EEG electrodes that cover the entire head. However, for real world applications, wearable devices that encompass just one (or a few) channels are desirable, which make the classification of EEG states even more challenging. With this as background, we applied variants of data driven Empirical Mode Decomposition (EMD) on McGurk EEG, which is an illusory perception of speech when the movement of lips does not match with the audio signal, for classifying whether the perception is affected by the visual cue or not. After applying a common pre-processing pipeline, we explored four EMD based frameworks to extract EEG features, which were classified using Random Forest. Among the four alternatives, the most effective framework decomposes the ensemble average of two classes of EEG into their respective intrinsic mode functions forming the basis on which the trials were projected to obtain features, which on classification resulted in accuracies of 63.66% using single electrode and 75.85% using three electrodes. The frequency band which plays vital role during audio-visual integration was also studied using traditional band pass filters. Of all, Gamma band was found to be the most prominent followed by alpha and beta bands which contemplates findings from previous studies.

Text summarization is the technique for generating a concise and precise summary of voluminous texts while focusing on the sections that convey useful information, and without losing the overall meaning. Although recent works are paying attentions in this domain, but still they have many limitations which need to be address.. This paper studies the attention model for abstractive text summarization and proposes three models named as Model 1, Model 2, and Model 3 which not only handle the limitations of the previous contemporary works but also strengthen the experimental results further. Empirical results on CNN/DailyMail dataset show that the proposed approach is promising.

The proposed model unites the robustness of the extractive and abstractive summarization strategies. Three tasks indispensable to automatic summarization, namely, apprehension, extraction, and abstraction, are performed by two specially designed networks, the highlighter RNN and the generator RNN. While the highlighter RNN collectively performs the task of highlighting and extraction for identifying the salient facts in the input text, the generator RNN fabricates the summary based on those facts. The summary is generated using word-level extraction with the help of term-frequency inverse document frequency (TFIDF) ranking factor. The union of the two strategies proves to surpass the ROUGE score results on the Gigaword dataset as compared to the simple abstractive approach for summarization.

With online social platforms becoming more and more accessible to the common masses, the volume of public utterances on a range of issues, events, and persons etc. has increased profoundly. Though most of the content is a manifestation of personal feelings of the individuals, yet a lot of this content often comprises of hate and offensive speech. Exchange of hate and offensive speech has now become a global phenomenon with increased intolerance among societies. However companies running these social media platforms need to discern and remove such unwanted content. This article focuses on automatic detection of hate and offensive speech from Twitter data by employing both conventional machine learning algorithms as well as deep learning architectures. We conducted extensive experiments on a benchmark 25K Twitter dataset with traditional machine learning algorithms as well as using deep learning architectures. The results obtained by us using deep learning architectures are better than state-of-the-art methods used for hate and offensive speech detection.

Natural Language Processing (NLP) and especially natural language text analysis have seen great advances in recent times. Usage of deep learning in text processing has revolutionized the techniques for text processing and achieved remarkable results. Different deep learning architectures like CNN, LSTM, and very recent Transformer have been used to achieve state of the art results variety on NLP tasks. In this work, we survey a host of deep learning architectures for text classification tasks. The work is specifically concerned with the classification of Hindi text. The research in the classification of morphologically rich and low resource Hindi language written in Devanagari script has been limited due to the absence of large labeled corpus. In this work, we used translated versions of English data-sets to evaluate models based on CNN, LSTM and Attention. Multilingual pre-trained sentence embeddings based on BERT and LASER are also compared to evaluate their effectiveness for the Hindi language. The paper also serves as a tutorial for popular text classification techniques.

Sentiment analysis is a crucial step in the social media data analysis. The majority of research works on sentiment analysis focus on sentiment polarity detection which identifies whether an input text is positive, negative or neutral. In this paper, we have implemented a stacked ensemble approach to sentiment polarity detection in Bengali tweets. The basic concept of stacked generalization is to fuse the outputs of the first level base classifiers using a second-level Meta classifier in an ensemble. In our ensemble method, we have used two types of base classifiers- multinomial Naïve Bayes classifiers and SVM that make use of a diverse set of features. Our proposed approach shows an improvement over some existing Bengali sentiment analysis approaches reported in the literature.

Natural Language Generation (NLG) is a crucial component of a Spoken Dialogue System. Its task is to generate utterances with intended attributes like fluency, variation, readability, scalability and adequacy. As the handcrafted models are rigid and tedious to build, people have proposed many statistical and deep-learning based models to bring about more suitable options for generating utterance on a given Dialogue-Act (DA). This paper presents some Recurrent Neural Network Language Generation (RNNLG) framework based models along with their analysis of how they extract intended meaning in terms of content planning (modelling semantic input) and surface realization (final sentence generation) on a proposed unaligned Hindi dataset. The models have shown consistent performance on our natively developed dataset where the Modified-Semantically-Controlled LSTM (MSC-LSTM) performs better than all in terms of total slot-error (T-Error).

Bengali is the sixth most popular spoken language in the world. Computerized detection of handwritten Bengali (Bangla Lekha) character is very difficult due to the diversity and veracity of characters. In this paper, we have proposed a modified state-of-the-art deep learning to tackle the problem of Bengali handwritten character recognition. This method used the lesser number of iterations to train than other comparable methods. The transfer learning on Resnet-50 deep convolutional neural network model is used on pretrained ImageNet dataset. One cycle policy is modified with varying the input image sizes to ensure faster training. Proposed method executed on BanglaLekha-Isolated dataset for evaluation that consists of 84 classes (50 Basic, 10 Numerals and 24 Compound Characters). We have achieved 97.12% accuracy in just 47 epochs. Proposed method gives very good results in terms of epoch and accuracy compare to other recent methods by considering number of classes. Without ensembling, proposed solution achieves state-of-the-art result and shows the effectiveness of ResNet-50 for classification of Bangla HCR.

Purchasing clothes that fit well on e-commerce portals can be problematic if consumers do not trust the fit of clothes based on specified size labels. This is especially a problem in developing countries, where size labels are not adequately standardized. In this paper, we introduce a system that can take a person’s mirror selfie as input and accurately predict anthropometric measurements using that image. These anthropometric measurements can then be used to predict clothing fit based on supplier-specific measurement-label mappings that are available or can easily be developed by e-commerce clothing retailers. The key novelty of our proposal is our use of mirror selfies, which physically ensures that an object of standardized and known size, a cellphone, is present in an image at a predictable orientation and location with the person being photographed. For predicting measurements, we experimented with a number of regression models. Empirical testing showed that the best regression models yield \({\le }5\%\) test set error with respect to 11 tailor-derived body measurements for each of 70 male subjects. The empirical success of our proposal leads us to believe that our proposed approach may considerably simplify the task of online body size prediction.

Automatic retinal vessel segmentation has turned out to be highly propitious for medical practitioners to diagnose diseases like glaucoma and diabetic retinopathy. These diseases are classified based on the thickness of the retinal vessel, the pressure imposed on the nerve endings and optical disc to cup ratio of the retina. The state-of-the-art device for this purpose presently available in the market is expensive and has scope to meliorate sensitivity and precision of its performance. Thus, automatic retinal blood vessel segmentation and classification is the need of the hour. In this paper, a novel non-local total variational retinex based retinal image preprocessing approach is proposed to extract the retinal vessel features and classify the vessels using ground truth images. Matlab implementation results indicate that an average accuracy of 94% with an acceptable range of sensitivity and specificity could be achieved on the retinal image database available online .

Cataract is one of the common problems among the humans. Cataract is the condition caused due to clouding of lens in the eye which eventually may lead to blindness. In last few years, data mining has been widely used to build the predictive model in various fields. In this paper, historical data of cataract patient has been used to build the predictive model. Random forest algorithm is one of the decision tree algorithms for predictive modeling. Random forest algorithm incorporates advantages of classification and regression. Present study uses random forest method to create a model for prediction of cataract. The random forest algorithm is also tested for Out of Bag estimation error.

Drowsiness of drivers is a critical problem and has recently attracted a lot of attention from both academia and industry. A real-time driver’s drowsiness detection system is often considered as a crucial component of an Advanced Driver Assistance System (ADAS). Although, there are a number of physical parameters associated with drowsiness like blink frequency, eye closure duration, pose, gaze, etc., yawing can also be used as an indicator of drowsiness. This work presents a novel deep learning-based framework for driver’s drowsiness prediction based on yawn detection in a video stream. The proposed approach uses a combination of a convolutional neural network (CNN), 1D-CNN, and bi-directional LSTM (Bi-LSTM). In the first step, the pipeline extracts the mouth region from each frame of the video using a combination of face and landmark detector. In the subsequent step, spatial information from the mouth region is extracted using a pre-trained deep convolutional neural network (DCNN). Finally, temporal information which models the evaluation of yawn using the extracted mouth feature is learned using a blend of 1D-CNN and bi-directional LSTM (Bi-LSTM). Experiments were performed on manually extracted and annotated video clips obtained from two publically available drowsiness detection dataset namely YawDD and NTHU-DDD. Experimental results show the effectiveness of the proposed approach both in terms of recognition accuracy and computational efficiency. Thus, the proposed pipeline is a good candidate for real-time implementation of yawn detection system for driver’s drowsiness prediction on an embedded device.

This paper concentrates on new technology development to support human mobility of ageing people. The development is based on human biomechanics. Movement of different joints and combination of joints are observed closely. On that basis, attempt has been made to design exoskeleton to support different body parts of old person. This research will help aged people for normal movement with ease. The paper aims to the major objectives which includes 1. To develop an exoskeleton powerful enough to transmit required high force to the muscles of the patients, so that patients don’t feel any difference between human physiotherapists & the exoskeleton. 2. To make the process of physiotherapy exercise beneficial for stroke patients. The proposed motor actuated wearable mechanical structure for upper limb (also known as upper limb exoskeleton) is to be integrated with virtual reality games. The elderly persons are noticed to have major problem in the lower limb i.e. hip, knee and ankle. The Paper introduces a measure to aid the people with lower limb disability.

For effective Human-Robot Interaction (HRI), an intelligent wheelchair (IW) need to be cognitively enhanced. Robot Operating System (ROS) has been steadily gaining popularity among robotics researchers as an open source framework for robot control. This paper presents ccaROS - a new ROS node for a Cognitive Collaborative Architecture to achieve better HRI for an IW. The design of the ROS node is presented. It provides mechanisms for obstacle avoidance, detection and adaption of user’s navigational strategy; seamless switching of driving control from machine to human and vice-versa. This would not only assist to achieve safe navigation but also allow retention of residual skills of the user. The effectiveness of ccaROS has been evaluated through simulation studies within a ROS-USARSim environment.

The efficient use of water is an issue that has captured the attention of scientists, technicians, and the community at large. The sustainability of water resources has been threatened by the current imbalance between water supply and demand. Intelligent consumption of water would contribute to the balance and reduce the waste in applications such as the agriculture. This paper shows the design of a water consumption monitoring system based on the Internet of Things (IoT). With the implementation of this system could be known in real time the consumption of water in a crop. In addition, the user of the system may take corrective actions that optimize their water consumption; this is achieved by applying the SEMMA methodology to evaluate the data obtained by the system using two cluster algorithms, Simple K-means and GenClus++. With the application of SEMMA it was possible to determine periods of water consumption that were considered as waste in the irrigation of crops, applying data analysis with both algorithms.

In the real world, entities are often connected via multiple relations, forming multi-relational network. These complex networks need novel models for their representation and sophisticated tools for their analysis. Community detection is one of the primary tools for the structural and functional analysis of the networks at the macroscopic level. Already a lot of research work has been done on discovering communities in the networks with only single relation. However, the research work on discovering communities in multi-relational network (MRN) is still in its early stages. In this article, we have proposed a novel approach to extract the communities in a multi-relational network using DeepWalk network embedding technique and Consensus clustering. Empirical study is conducted on the real-world publicly available Twitter datasets. In our observations we found that our proposed model performs significantly better than some of the baseline approaches based on spectral clustering algorithm, modularity maximization, block clustering and non-negative matrix factorization.

The type of training in virtual-reality (VR) environment plays a crucial role in enhancing military personnel’s decision-making ability. Little is currently known about how exposure to different types of training in VR designs may assist operators in getting trained in different simulated scenarios. We developed a VR search-and-shoot simulation with two scenarios in task complexity (novice and professional). Thirty healthy subjects played both the novice and professional scenarios in the VR design. Half of the participants were given novice training first, and half of the participants were given professional training first. We took various cognitive and behavioral measures into consideration for statistical analyses. Results disclosed that the participants who faced the professional scenario first fared better than the participants who faced the novice scenario first. We discuss the implication of our results involving VR technologies for creating effective environments for training military personnel.

This article shows the experience in the implementation of a tool called Glyph Reader, which is an application that has two interfaces, Web and Mobile and that responds to the need for an educational and interactive resource whose main objective is the Multisensory stimulation for literacy training in a population with cognitive disabilities and/specific learning disorder. The design of the activities that this application has is based on the theoretical model of multisensory stimulation Orton Gillingham, which seeks the development of basic skills for decoding isolated words based on a phonetic - graphic analysis of them. The techniques within this model use the basic concepts of intersensory integration of simultaneous visual-auditory-kinesthetic- tactile differentiation (VAKT), to which the Glyph Reader application takes full advantage, by including graphic phonetic recognition and training activities of syllables/words (exercises with symphons and exercises with combinations of consonants or working syllables), which pass from basic levels to complex levels of decoding, necessary for the development of literacy skills. The study sample for software validation is 250 students from the Eustorgio Salgar educational institution, in the municipality of Puerto Colombia, in the department of Atlántico - Colombia

This article shows the construction of software applications Cyclon Language First Grade App, like a strategy in which communities of practice, learning, knowledge, innovation and transformation are generated, understood as a transversal process, where collaborative, problematizing learning is encouraged, by critical inquiry, permanent interaction, cultural negotiations and the dialogue of knowledge, typical of the pedagogical proposal of the Ondas program. It is summarized in the following aspects: “Building an identity that incorporates the recognition of science and technology as a constituent element of everyday culture both in individuals and in the communities and institutions of which they are part, involving various sectors of society: productive, social, political, state and in the various territorial areas: local, departmental and national. Development of forms of organization oriented to the appropriation of values that recognize a cultural identity around science and technology in the aspects mentioned in the previous point. This implies models of participation, social mobilization and public recognition of scientific and technological activity. On the other hand, the incorporation of the research activity in the elementary and middle school involves the development of national, departmental and local financing mechanisms; in such a way that children and young people can develop their abilities and talents in a favorable environment of both social recognition and economic conditions. Development of a methodological strategy supported by ICT that helps the Colombian population to recognize and apply both individually and collectively, science and technology through research activities designed according to the characteristics of the scientific method. “The appropriation of ICTs as a constitutive part of the citizen and democratic culture of the CT + I and the construction of virtual reality as central to the process of knowledge democratization.

Since there has been an important increase in unmanned vehicles systems research such as quadrotors, a mathematical model and PID control laws are studied. Based on some dynamic variables, PID control is applied to compute a controller to be then use in autopilot simulations. As this kind of VTOL vehicle seems to be unstable, the aim of this work is to change even other flight mechanics parameters and control gains to study attitude and altitude variations. A well-known computational tool is used for simulation purposes, performance analysis and validation.