Current wearable health monitoring ecosystems typically operate in isolation, generating fragmented physiological streams that lack semantic interoperability. While commercial devices track individual metrics such as pulse rate or sleep duration, they fail to model the complex, temporal interdependencies between these physiological states. In this work, we propose the Multimodal Health state Monitoring System (MHMS), a unified framework designed to improve the interpretation of multimodal sensor data. The system aggregates data through networks from consumer-grade wearables to derive five health parameters of sleep quality, psychosomatic stress, cardiovascular health, cognitive load, and emotional vitality. A deep learning method adopting the spatio-temporal graph attention transformer technique treats these parameters as interconnected nodes in a temporal graph. This enables the system to model directional temporal dependencies by learning to forecast health trajectories. Experimental validation using standard datasets of WESAD and DREAMT demonstrates a sleep stage classification accuracy of 88%, a stress detection accuracy of 95.26%, and an R-square score of 0.82 in predictive modeling. The proposed solution offers a potential direction for a future e-health monitoring system in compliance with Recommendation ITU-T H.810.

Video analytics faces complex challenges in object detection and classification. Deep learning based approaches have achieved remarkable success in past decade. However, existing object identification models that utilize backbone’s core features still present challenges due to their lack of semantic information. To address these issues, a novel object detection and classification framework utilizing Batch normalization and Softswish activation adapted ResNet (BS2ResNet) and Logistic Tanh Kaiming Bi-directional Long Short Term Memory (LTK-Bi-LSTM) techniques was proposed. The framework employs frame conversion, noise removal, and contrast elevation during frame pre-processing, followed by background subtraction using the Supreme Distance-centered Fuzzy C-Means (SD-FCM) clustering algorithm, and edge detection using the Hyperbolic Tangent Kernel Canny Edge Detector (HTKCED). BS2ResNet is then used for object detection, and features are extracted and passed to the LTK-Bi-LSTM neural network for object classification. The proposed system was found to improve object detection and classification accuracy, outperforming existing techniques.

The Sequential Pattern Mining (SPM) is a challenging task in data streams due to huge memory and computational costs to meet accuracy in mined results. Sequential patterns mined from target stream in traditional batch-based processing results in pattern loss when the batches are processed independently, where the pattern frequency is determined local to the batch. However, if a pattern is frequent in the stream and its items appear in various batches, then this pattern never becomes frequent and hence requires pruning. To address this issue, the sequences are clustered by similarity using Adaptive Hierarchical Clustering (AHC) and Batch-Free Top-K Sequential Pattern Mining (BFTKSPM) algorithms proposed to mine approximate sequential patterns over data streams. The BFTKSPM algorithm targets data stream in a continuous and batch-free manner. The top-k sequential patterns are extracted from data streams and are maintained in an inverted tree structure. The experimental results of the proposed algorithm are carried out on benchmark datasets for data streams and it outperforms the existing batch-based methods in terms of execution time, memory, precision, recall, and F1-score.

Elderly healthcare needs to provide solutions to the complex challenges of real-time tracking, monitoring, and health anomaly detection. The proposed system integrates voice and video-based health anomaly detection to provide continuous monitoring and personalized care to elderly individuals in a home setting. It includes emotion recognition in voice and video-based posture identification. The voice module is developed using convolutional neural network (CNN) and long short-term memory (LSTM) deep learning models, while a transfer learning technique is used in the implementation of the video module. The system is implemented on the Raspberry Pi platform to ensure a cost-effective solution. Attention mechanisms are incorporated into machine learning models to enhance accuracy, particularly for fall detection. Validation experiments with the Raspberry Pi implementation achieved an accuracy of 96.34% for the voice module and 87.91% for the video module on standard datasets for elderly care. This method aligns with the ITU standardization efforts.

A Multiple moving object detection, tracking, and counting algorithm is mainly designed exclusively suitable for congested areas. The counting system can alleviate the betrayal performance in the crowded areas. Most of the existing methods developed for tracking and counting face serious challenges in detection due to high densities of the target. This condition urged the researchers to update the existing systems. The present methodology was designed to address such issues. In the present methodology, the contrast was initially enhanced between the objects and their backgrounds using a Double Plateau Histogram Equalization (DPHE). Then, the motion was estimated for the contrast-enhanced image to identify the moment of the object using the modified Adaptive Distance Covariance Rood Pattern Search (ADCRPS) algorithm. After that, the morphological operation was deployed to sharpen the images by removing all the unwanted things. Then, the features were extracted and important features were selected using the modified Chaotic Tent Shuffled Shepherd Optimization (CTSSO) Algorithm. With the selected features object, detection was done using the proposed Scaled Non-Monotonic Cauchy Dense Convolutional Neural Network (SNMC-DenCNN). The detected object was then tracked with the aid of Channel and Spatial Reliability Tracker (CSRT). Finally, the objects were counted by intersection over union (IOU) by explicitly computing the association between detected and tracked objects. Also, the experimental results showed the effectiveness and efficiency of the proposed system with enhanced accuracy.

We model the anti-jamming system for cognitive radio networks under multiple jammer attacks using game theory as incomplete information games. The game model has three players: nature (representing channel impairment), jammer, and cognitive radio with sequential moves. Typically, sequential games with incomplete information use the Perfect Bayesian equilibrium solution concept, where beliefs are determined by Bayes' rule to identify the player type. In the proposed Perfect Hidden Markov -Viterbi equilibrium solution concept, beliefs are determined by the Hidden Markov Model - Viterbi decoding instead of Bayes' rule to identify the jammer type. A general form of expression is derived to compute players' payoff for probability-based belief systems to find equilibrium in anti-jamming games with incomplete information under multiple jammer attacks.The simulation results show that over time, with a gradual reduction in the bit error rate at the receiver without any channel coding for Binary Phase Shift Keying (BPSK) modulation and in the noisy Rayleigh fading environment, the proposed Perfect Hidden Markov - Viterbi equilibrium solution concept performs better than Perfect Bayesian equilibriumby 0.8 % and the traditional pseudorandom-based frequency hopping technique by 1 % in terms of the average uncoded bit error rate.

An efficient model to detect and track the objects in adverse weather is proposed using Tanh Softmax (TSM) EfficientDet and Jaccard Similarity based Kuhn-Munkres (JS-KM) with Pearson-Retinex in this paper. The noises were initially removed using Differential Log Energy Entropy adapted Wiener Filter (DLE-WF). The Log Energy Entropy value was calculated between the pixels instead of calculating the local mean of a pixel in the normal Wiener filter. Also, the segmentation technique was carried out using Fringe Binarization adapted K-Means Algorithm (FBKMA). The movement of segmented objects was detected using the optical flow technique, in which the optical flow was computed using the Horn-Schunck algorithm. After motion estimation, the final step in the proposed system is object tracking. The motion-estimated objects were treated as the target that is initially in the first frame. The target was tracked by JS-KM algorithm in the subsequent frame. At last, the experiential evaluation is conducted to confirm the proposed model’s efficacy. The outcomes of Detection in Adverse Weather Nature (DAWN) dataset proved that in comparison to the prevailing models, a better performance was achieved by the proposed methodology.

Pedestrian navigation in an urban environment based on traditional digital mapping systems is constrained by the inherent limitations of existing online mapping services. The major challenges include the reliance on global positioning system (GPS), whose signals are not received well in some locations, and the inferior user experience caused by the lack of information about unknown surroundings. This article describes the design and development of a markerless augmented reality-based pedestrian navigation system that can handle navigation even in the absence of GPS signals. It improves the user experience by providing a novel landmark recognition feature, which allows users to identify nearby buildings or streets during navigation. To mitigate the absence of GPS signals, a novel user localization method utilizing a step count-based distance estimator is proposed. The experimental results show the location accuracy of 2.3 meters on average and the increase in step count detection accuracy of nearly 0.5% in comparison with existing technologies available in smartwatches, and an average latency of 74 milliseconds in system response in an urban environment. The proposed solution can be used as a mobile application on smartphones and has the potential to contribute to the smart city-related standardization activities of ITU-T Study Group 16.

The elderly healthcare requires an innovative approach to address multifaceted challenges in tracking, monitoring, and reporting in real-time. The proposed solution harnesses the capabilities of voice and video-based anomaly detection systems to offer continuous monitoring, personalized support, and timely intervention for the physical and emotional well-being of elderly individuals. Central to the proposed system is the integration of real-time voice emotion recognition and video-based posture recognition modules, constructed using cutting-edge deep learning and transfer learning models respectively. These modules are deployed on the Raspberry Pi platform, ensuring accessibility and efficiency. Moreover, attention mechanisms are incorporated to boost accuracy and effectiveness in detecting health anomalies, with a particular focus on identifying falls. The proposed elderly companion system implemented on Raspberry Pi achieves a validation accuracy of 96.34% in voice module and 87.91% in video module in delivering comprehensive healthcare for the elderly. The proposed solution demonstrates a potential work for standardization through the ITU/WHO Focus Group on Al for Health (FG-AI4H).

Smart home automation systems require convenient and efficient user interface to control home appliances. Gesture recognition-based solutions offer flexibility to the users and play a crucial role in advancing human-computer interaction and immersive computing environments. This work proposes a novel solution leveraging deep learning techniques with attention mechanisms including self-attention tailored for processing 3D tensors derived from the gesture images. A set of hand gestures is defined, and the system is trained and optimized to meet the real time requirements in controlling devices. To improve the accuracy, the model is parallelly trained with dynamic learning to adaptively fuse with the classification module. The proposed modular architecture is implemented using Raspberry Pi with IoT devices for a typical home environment. The test result achieves gesture classification accuracy of98.24% and latency of about 0.2 seconds in real time control. The working model highlights a practical solution under ITU-T Recommendation J.1611 which deals with the functional requirements of a smart home and gateway.

Object detection and classification are the trending research topics in the field of computer vision because of their applications like visual surveillance. However, the vision-based objects detection and classification methods still suffer from detecting smaller objects and dense objects in the complex dynamic environment with high accuracy and precision. The present paper proposes a novel enhanced method to detect and classify objects using Hyperbolic Tangent based You Only Look Once V4 with a Modified Manta-Ray Foraging Optimization-based Convolution Neural Network. Initially, in the pre-processing, the video data was converted into image sequences and Polynomial Adaptive Edge was applied to preserve the Algorithm method for image resizing and noise removal. The noiseless resized image sequences contrast was enhanced using Contrast Limited Adaptive Edge Preserving Algorithm. And, with the contrast-enhanced image sequences, the Hyperbolic Tangent based You Only Look Once V4 was trained for object detection. Additionally, to detect smaller objects with high accuracy, Grasp configuration was observed for every detected object. Finally, the Modified Manta-Ray Foraging Optimization-based Convolution Neural Network method was carried out for the detection and the classification of objects. Comparative experiments were conducted on various benchmark datasets and methods that showed improved accurate detection and classification results.

An episode rule mining to extract useful and important patterns or episodes from large event sequences represents the temporal implication of associating the antecedent and consequent episodes. The existing technique for mining precise-positioning episode rules from event sequences, mines serial episodes resulting in enormous memory consumption. To resolve this issue, the proposed work ensures the generation of fixed-gap episodes and parameter settings through the use of Particle Swarm Optimization mechanism. Fixed-gap episodes are generated using Natural Exponent Inertia Weight-based Particle Swarm Optimization algorithm. In this paper, a new technique called Mining Serial Episode Rules (MSER) is proposed, which utilizes the correlation between episodes and the generation of parameter selection where the occurrence time of an event is specified in the consequent. Further, a trie-based data structure to mine MSER along with a pruning technique is incorporated in the proposed methodology to improve the performance. The efficiency of the proposed algorithm MSER is evaluated on three benchmark data sets Retail, Kosarak, and MSNBC where the experimental results outperform the existing methods with respect to memory usage and execution time.

On grounds of the advent of real-time applications, like autonomous driving, visual surveillance, and sports analysis, there is an augmenting focus of attention towards Multiple-Object Tracking (MOT). The tracking-by-detection paradigm, a commonly utilized approach, connects the existing recognition hypotheses to the formerly assessed object trajectories by comparing the similarities of the appearance or the motion between them. For an efficient detection and tracking of the numerous objects in a complex environment, a Pearson Similarity-centred Kuhn-Munkres (PS-KM) algorithm was proposed in the present study. In this light, the input videos were, initially, gathered from the MOT dataset and converted into frames. The background subtraction occurred which filtered the inappropriate data concerning the frames after the frame conversion stage. Then, the extraction of features from the frames was executed. Afterwards, the higher dimensional features were transformed into lower-dimensional features, and feature reduction process was performed with the aid of Information Gain-centred Singular Value Decomposition (IG-SVD). Next, using the Modified Recurrent Neural Network (MRNN) method, classification was executed which identified the categories of the objects additionally. The PS-KM algorithm identified that the recognized objects were tracked. Finally, the experimental outcomes exhibited that numerous targets were precisely tracked by the proposed system with 97% accuracy with a low false positive rate (FPR) of 2.3%. It was also proved that the present techniques viz. RNN, CNN, and KNN, were effective with regard to the existing models.

A large event sequence can generate episode rules that are patterns which help to identify the possible dependencies existing among event types. Frequent episodes occurring in a simple sequence of events are commonly used for mining the episodes from a sequential database. Mining serial positioning episode rules (MSPER) using a fixed-gap episode occurrence suffers from unsatisfied scalability with complex sequences to test whether an episode occurs in a sequence. Large number of redundant nodes was generated in the MSPER-trie-based data structure. In this paper, forward and backward search algorithm (FBSA) is proposed here to detect minimal occurrences of frequent peak episodes. An extensive correlation of parameter settings and the generating procedure of fixed-gap episodes are carried out. To generate a fixed-gap episode and estimate the variance that decides the parameter selection in event sequences, Spearman's correlation coefficient is used for verifying the sequence of occurrences of the episodes. MFSPER with FBSA is developed to eliminate the frequent sequence scans and redundant event sets. The MFSPER-FBSA stores the minimal occurrences of frequent peak episodes from the event sequences. The experimental evaluation on benchmark datasets shows that the proposed technique outperforms the existing methods with respect to memory, execution time, recall and precision.

In image captioning improving the content of the image by describing the meaning of the picture is a challenge as it should not only be understandable to the user but also described in a short and clear sentence. The proposed solution uses CNN and LSTM as captioning models in an Encoder and the Decoder methodology is used to translate an image into a sentence. The Xception architecture is modified by adding a depth wise convolution layer. A custom activation function is created based on swish and mish. CNN is used for feature extraction, RNN is used for sequence prediction, and LSTM for framing the words into a sentence. The proposed work is validated on two-dimensional image datasets such as dog category data extracted from the flicker8k dataset and real-time images captured through a webcam. The training/testing shows improved loss value, caption prediction time, and an increase in the quality of caption in terms of the BLEU@I parameter.

Pedestrian navigation using traditional mapping systems is constrained by the inherent limitations of existing digital online mapping services. The major challenges include complete reliance on GPS for user localization and inferior user experience caused by lack of information about the surroundings, especially in unknown environments. In this paper, we design and develop a marker-less augmented reality-based pedestrian navigation system which can handle navigation even in the absence of GPS as well as improve user experience by providing a novel landmark recognition feature, which allows users to identify nearby buildings or streets during navigation. To mitigate the absence of a GPS signal, a user localization method utilizing a step count-based distance estimator is proposed. The performance comparison with existing state of the art techniques and devices shows locational accuracy of 2.5 meters on average and a step count detection accuracy increase of nearly 0.5% with a latency of 70 milliseconds in an urban environment. The proposed solution is intended to be used as a mobile application on smartphones and has a potential to contribute to the smart city-related standardization activities of ITU-T Study Group 16.

Human activity recognition has emerged as a challenging research domain for video analysis. The major issue for abnormal activity recognition in a streaming video is the presence of the large spatio-temporal data along with the constraints of communication networks affecting the quality of received data for analysis. In this paper, we propose a deep learning-based system to identify abnormal human activities using a combination of Skeleton Activity Forecasting (SAF) and a Bi-LSTM network. The generated skeleton joint points of a human subject are used for the pose estimation. The skeleton tracking and regions of interest points are estimated on a streaming video from an IP networked camera. The extracted interest points and their corresponding features are optimized and used to classify them as normal, abnormal or suspicious actions. The proposed system complies with Recommendation ITU-T H.627 “Signalling and protocols for a video surveillance system” and has been experimented and evaluated over benchmarked data sets for the recognition of human actions. The system performance attains a precision of 85.6% and an accuracy of 97.2% in recognizing different actions.

The Frequent Episode Mining (FEM) is a challenging framework to identify frequent episodes from a sequence database. In a sequence, an ordered collection of events defines an episode, and frequent episodes are only considered by the earlier studies. Also, it doesn’t support for the serial based episode rule mining. In this work, the episode rules are mined with precise and serial based rule mining considering the temporal factor, so that, the occurrence time of the consequent is specified in contrast to the traditional episode rule mining. The proposed work has a larger number of candidates and specific time constraints to generate the fixed-gap episodes, and mining such episodes from whole sequence where the time span between any two events is a constant which is utilized to improve the proposed framework’s performance. In order to improve the efficiency, an Optimal Fixed-gap Episode Occurrence (OFEO) is performed using the Natural Exponent Inertia Weight based Swallow Swarm Optimization (NEIWSSO) algorithm. The temporal constraints significantly evaluate the effectiveness of episode mining, and a noticeable advantage of the present work is to generate optimal fixed-gap episodes for better prediction. The effective use of memory consumption and performance enhancement is achieved by developing new trie-based data structure for Mining Serial Positioning Episode Rules (MSPER) using a pruning method. The position of frequent events is updated in the precise-positioning episode rule trie instead of frequent events to reduce the memory space. The benchmark datasets Retail, Kosarak, and MSNBC is used to evaluate the proposed algorithm’s efficiency. Eventually, it is found that it outperforms the existing techniques with respect to memory consumption and execution time. On an average, the proposed algorithm achieves 28 times lesser execution time and consumes 45.5% less memory space for the highest minimum support value on the Retail dataset compared to existing methods.

Home interior design is often a challenging and time-consuming task due to several mismatches between the selection of interiors in a shop and their composition in a target room. One of these mismatches is a lack of knowledge in relation to the target room and to other interiors when an item is seen in a shop. In this paper, we propose to design and develop a system model by which users can virtually try out various settings of home interiors by using marker-less Augmented Reality (AR). To minimize the latency of the system, we apply the hybrid approach of combining an AR framework with a Simultaneous Localization And Mapping (SLAM) algorithm, where the 3D feature points are updated dynamically, thereby avoiding occlusion when multiple interiors are placed on the real-world view. Dimension scanning is deployed to provide accurate live-size scaling to the real-world environment whereas transformations are handled efficiently by means of a touch gesture recognition. The proposed system model for home interior design can be easily implemented as a mobile application. The performance evaluation through experiments shows that with the SLAM algorithm, the average response time of the proposed system improves by 43%. The proposed system is intended to meet the requirements of virtual and augmented reality services specified by ITU-T Study Group 16.