Abstract Sign language recognition is one of the fundamental ways to assist deaf people to communicate with others. An accurate vision-based sign language recognition system using deep learning is a fundamental goal for many researchers. Deep convolutional neural networks have been extensively considered in the last few years, and a slew of architectures have been proposed. Recently, Vision Transformer and other Transformers have shown apparent advantages in object recognition compared to traditional computer vision models such as Faster R-CNN, YOLO, SSD, and other deep learning models. In this paper, we propose a Vision Transformer-based sign language recognition method called DETR (Detection Transformer), aiming to improve the current state-of-the-art sign language recognition accuracy. The DETR method proposed in this paper is able to recognize sign language from digital videos with a high accuracy using a new deep learning model ResNet152 + FPN (i.e., Feature Pyramid Network), which is based on Detection Transformer. Our experiments show that the method has excellent potential for improving sign language recognition accuracy. For instance, our newly proposed net ResNet152 + FPN is able to enhance the detection accuracy up to 1.70% on the test dataset of sign language compared to the standard Detection Transformer models. Besides, an overall accuracy 96.45% was attained by using the proposed method.

Dementia patients living alone in the communities without much support from near and dear ones find it chal-lenging to receive instant help when needed including during emergencies. This work focuses on designing an end - to-end response system primarily to support early-stage Alzheimer's dementia (AD) patients living alone in their homes, in case of emergencies, including medical emergencies. The system aided with pervasive technologies can integrate AD patients, informal caregivers, and formal caregivers to support AD patients in need. Informal caregivers act as first responders to attend to patients and selecting appropriate informal caregivers based on certain predefined parameters is an important component of this system. This work has used single and ensemble Machine Learning (ML) algorithms for binary (to check if informal caregiver is available) and multiclass (to select the most suitable informal caregiver) classification.

Globally, an increasing number of elderly people live alone at home either by choice or out of compulsion, and many have health issues. During a medical emergency involving an elderly person living alone, getting prompt access to formal medical aids like calling an ambulance or paramedics can be critical. Moreover, an ambulance can also arrive late adding to the patient's suffering. In such situations, informal caregivers can act as first responders to attend to the patients even before an ambulance arrives. The selection of appropriate informal caregivers to attend a medical emergency incident remains a key issue and this paper proposes an informal caregiver selection process based on certain pre-defined characteristics of informal caregivers. Single and ensemble multiclass classification algorithms are considered in this work to rank informal caregivers and preliminary simulation results have shown that ensemble models perform better than single models.

The increasing demand for higher data rates pushes the boundaries of the currently used radio spectrum. The terahertz (THz) frequency band (0.1-10 THz) is widely considered by scientific community to address spectrum scarcity. In this paper we developed a theoretical model for device-to-device (D2D) communication operating in THz band. We also derived a close form formula of data rates, outage probability, and energy efficiency. Our simulation results show an improvement of data rates and energy efficiency while decreasing the outage probability of D2D communication in THz. For instance, there is 86% of increase in energy efficiency when the transmission power is 19dBm. Finally, the improvement of energy efficiency is 87% using optimal transmission power due to 50 resource blocks.

Natural disasters, such as earthquakes, can cause severe destruction and create havoc in the society. Buildings and other structures may collapse during disaster incidents causing injuries and deaths to victims trapped under debris and rubble. Immediately after a natural disaster incident, it becomes extremely difficult for first responders and rescuers to find and save trapped victims. Often searches are carried out blindly in random locations, which delay the rescue of the victims. This paper introduces a Smartphone Assisted Disaster Recovery (SmartDR) method for post-disaster communication using Smartphones. SmartDR utilizes the device-to-device (D2D) communication technology in Fifth Generation (5G) networks, which enables direct communication between proximate devices without the need of relaying through a network infrastructure, such as mobile access points or mobile base stations. We examine a scenario of multi-hop D2D communication where smartphones carried by trapped victims and other people in disaster affected areas can self-detect the occurrence of a disaster incident by monitoring the radio environment and then can self-switch to a disaster mode to transmit emergency help messages with their location coordinates to other nearby smartphones. To locate other nearby smartphones also operating in the disaster mode and in the same channel, each smartphone runs a rendezvous process. The emergency messages are thus relayed to the functional base station or rescue centre. To facilitate routing of the emergency messages, we propose a path selection algorithm, which considers both delay and the leftover energy of a device (a smartphone in this case). Thus, the SmartDR method includes: (i) a multi-channel channel hopping rendezvous protocol to improve the victim localization or neighbour discovery, and (ii) an energy-aware multi-path routing (Energy-aware ad-hoc on-demand distance vector or E-AODV) protocol to overcome the higher energy depletion rate at devices associated with single shortest path routing. The SmartDR method can guide search and rescue operations and increase the possibility of saving lives immediately aftermath a disaster incident. A simulation-based performance study is conducted to evaluate the protocol performance in post-disaster scenario. Simulation results show that a significant performance gain is achievable when a device utilizes the channel information for the rendezvous process and the leftover energy for routing path selection. Our results show that peer discovery in multi-channel D2D environment can be significantly improved when channel quality information is considered in CH sequence design. Moreover, selecting a routing path based on LoE of a device and the standard deviation of residual energy of a path, can not only enhance the network lifetime, but also reduce the chance of network being partitioned.

In distributed cognitive radio (CR) networks, rendezvous (RDV) is one of the most critical issues. Due to the dynamic radio environment, achieving RDV on a predetermined common control channel (CCC) is a challenging task. Channel hopping (CH) provides an effective method to guarantee RDV in cognitive radio ad-hoc networks (CRAHNs). Most of the existing CH schemes utilize the channel quantity as an input to the family of mathematical concepts such as prime number theory, Chinese remainder theory (CRT), quorum system and combinatorial block design to achieve RDV. However, RDV on a channel is rather influenced by channel quality or CR user's preference on which it wants to achieve RDV on available channels. Based on this philosophy, a channel rank based torus quorum CH RDV protocol (TQCH) is proposed which finds a commonly available channel between a pair of CR nodes. We formulate the channel ranking as a linear optimisation problem based on the channel availability under collision constraints. A detailed mathematical formulation is derived to estimate the degree of overlap in terms of expected quorum overlap size. Simulation results show that the proposed TQCH scheme outperforms than that of other CH schemes in terms of average time-to-rendezvous (ATTR) and the degree of overlap in asymmetric channel scenario.

Rendezvous in cognitive radio ad-hoc networks (CRAHNs) is a key step for a pair of unknown cognitive radio (CR) users to initiate communication. Channel hopping (CH) provides an effective method to guarantee rendezvous in CRAHNs. To design a CH scheme, assumption of symmetric channel information is widely used in the literature. This assumption may ease the CH design, but unable to capture the dynamic radio environment. To achieve the rendezvous in the shortest time, most of the existing CH sequences utilized the family of mathematical concepts such as prime number theory, Chinese remainder theory (CRT), quorum system and combinatorial block design and so on. However, rendezvous on a channel is rather influenced by the CR user preference on which it wants to achieve rendezvous. In this paper we address rendezvous problem with asymmetric channel information. This paper proposes an adaptive CH sequence based on local channel sensing information to achieve rendezvous in finite time and guarantees overlap on all available common channels. Results obtained show that variation in channel ranking has a significant influence on CH performance.

Next generation wireless networks will be heterogeneous, where several primary users (PU e.g. licensed users) and secondary users (SU e.g. unlicensed users) can operate in the same dynamic and reconfigurable networks at a given time. The major challenge in this heterogeneous radio environment is to enable the coexistence between PU and SU which will further improve the efficient use of radio spectrum. Most of the existing coexistence techniques encounter with challenges due to lack of a priori knowledge about the primary system. Therefore Cognitive pilot channel (CPC) is a proposed approach which could enhance the coexistence by conveying some priori information. However, to achieve a peaceful coexistence it is essential to adopt a mitigation technique according to the CPC information. There is no algorithm has been described so far to integrate the CPC information with existing mitigation technique. In this paper, we proposed a novel power adaptation and integrated zone model (PAIZM) CPC algorithm for peaceful coexistence in heterogeneous networks. Moreover we have implemented and evaluated the PAIZM-CPC model as a coexistence enabler. The results show an enhancement compared with the existing coexistence techniques.