IFIP Open Access Digital Library
Not a member yet
22614 research outputs found
Sort by
Realistic Avatar Control Through Video-Driven Animation for Augmented Reality
No full textPart 2: Applications of AI/ML in Image ProcessingInternational audienceThis paper proposes an efficient real-time framework to generate detailed avatar animations solely from monocular camera videos, avoiding costly motion capture equipment. It extracts 3D facial and body landmarks using Blaze Pose key points on the input video. A novel adaptor mapping function then transforms the 2D landmark topology into diverse 3D avatar rigs, enabling the animation of different characters. The unified approach produces high-fidelity lip sync, expressions, gestures, and full-body motions in real-time. Extensive experiments demonstrate the framework generates realistic avatar mimicry of humans in video for immersive real-time applications in VR/AR entertainment and animation. A novel adaptor mapping function transforms 2D landmarks extracted by Blaze Pose into diverse 3D avatar rigs, overcoming topology limitations. The unified approach produces detailed facial expressions, lip sync, gestures, and body motions in real-time, enabling the avatar to mimic humans in video. Extensive experiments validate that the framework generates realistic avatar animations comparable to motion capture, with applications in real-time VR/AR. Key innovations include the novel mapping function to transform 2D landmarks into 3D avatar motions, and the real-time performance to animate avatars that closely imitate people in monocular video
Drug Sentiment Analysis: A Comprehensive Study Using Regression Models and Natural Language Processing
No full textPart 1: Applications of AI/ML in Natural Language ProcessingInternational audienceSentiment analysis is a critical technology in the era of ubiquitous digital communication that helps extract user opinions from large volumes of textual data. Our approach combines the Random Forest, XGBoost, and linear regression models with sentiment analysis and feature engineering via Term Frequency-Inverse Document Frequency (TF-IDF). The study provides a comprehensive understanding of drug-related sentiments by utilizing a diverse dataset that includes user-generated reviews and associated metadata. Moreover, the examination integrates medication-specific attributes and normalized counts, augmenting the level of detail in sentiment insights. Our technological approach shows effectiveness in capturing complex emotions present in online drug discussion user discourse. This study tackles the challenging task of extracting sentiments from drug-related online discussions while also contributing to the growing field of sentiment analysis. The results clarify how sophisticated regression models can be used to extract minute sentiment nuances, setting the stage for further advancements in pharmaceutical informatics. This paper serves as a comprehensive guide for researchers, providing invaluable insights into the utilization of advanced methodologies in sentiment analysis within the realm of pharmaceuticals
Driver’s Distraction Detection via Hybrid CNN-LSTM
No full textPart 2: Applications of AI/ML in Image ProcessingInternational audienceThis study introduces a novel hybrid architecture consisting of a Convolutional Neural Network (CNN) and Long Short-Term Memory (LSTM) to tackle the pressing problem of inattentive driving. Our approach leverages the temporal modeling capabilities of LSTMs and the spatial feature extraction capabilities of CNNs to perform real-time detection and analysis of various diversions, including but not limited to dining, phone calls, and messaging. Subsequent to undergoing training on an extensive dataset, the hybrid model attains an unprecedented level of accuracy, thereby promoting road safety for all. A convolutional neural network (CNN) processes visual input data while a long short-term memory (LSTM) component accumulates temporal correlations with the purpose of identifying and categorizing disruptive actions. Audio signals may assist motorists in refocusing their attention and mitigating potential dangers. This study not only signifies a substantial advancement in the domain of diversion detection but also holds promising implications for the progression of autonomous vehicles and advanced driver assistance systems (ADAS)
Rule-Based Control Algorithm to Explore Energy Flexibility from Residential Pool Filtration Pumps
No full textPart 5: Energy Management and SustainabilityInternational audienceResidential pool filtration pumps, together with the HVAC systems, represent the highest residential consumption loads. The efficiency of pool filtration pumps has been a recurring topic of study; however, the same cannot be said for the optimization of their operation. This study contributes to the scientific community by developing a specific control algorithm for filtration pumps, filling some gaps in the existing literature. The research addresses the energy flexibility of residential pool filtration pumps, introducing a rule-based control algorithm to reduce associated energy costs by utilizing surplus solar energy from a residential photovoltaic system. The results, collected from an illustrative case study, demonstrate significant reductions in pump operation costs (−46%) and total electricity bill (−18%). These results highlight the potential of the proposed algorithm to improve the operation of filtration pumps, contributing to increase the self-consumption of on-site generation and to reduce electricity costs
Multimodal Creativity State Detection from Speech and Voice
No full textPart 2: Human-Robot CollaborationInternational audienceModern society has been shaped and advanced by creativity, which has an impact on many facets of human expression and interaction. This research explores the intersection of technology and human interaction by proposing a computational multimodal creativity state identification from speech and voice. This study examines the relationship between language elements and emotion detection, as well as the emotional dimensions (arousal and valence), with a focus on the dynamic and process of identifying creative states through the analysis of speech and voice modalities. To achieve the creativity state detection, it uses a CNN model that was trained using four datasets: Crema-D, Ravdess, Savee, and Tess. A linguistic prompt creativity test is included in the study to provide external validation of the suggested model. The findings show a strong relationship between emotion and emotional dimension, alongside the machine-based detection methods
Innovative Digital Forensic and Investigation Tools for Law Enforcement: The EMPOWER & TRACY Approach
No full textPart 1: The 9th Workshop on “5G – Putting Intelligence to the Network Edge” (5G-PINE)International audienceConstantly growing digitalisation in all sectors and the rapidly changing technological landscape provide vast opportunities for criminals and terrorists. Law Enforcement Agencies (LEAs) often lack the necessary technical and financial means as well as digital skills when preventing, detecting, investigating or prosecuting criminal and terrorist activities supported by advanced technologies. The overall goal of EMPOWER is to foster the uptake of innovative solutions based upon AI powered tools allowing Law Enforcement Agencies (LEAs) to increase their capabilities in such investigative fields. To that end, EMPOWER will pilot test a total of eight investigative tools in the fields of Image/Video, Voice/Text and Federated Learning. During the project, eight tools will be brought to Technology Readiness Level 8, following their testing by two LEAs with operational datasets in real-life environments. The TRACY solution is an open-source platform with the aim of up taking an AI-based system, by running large scale pilots on LEA’s premises, using telecommunications metadata in a fully operational environment, in full respect of fundamental rights and personal data protection. For greater impact, the solution shall be validated by additional LEAs within the project, with the aim to be permanently used after its completion
On 6G-Enabled SDN-Based Mobile Network User Plane with DRL-Based Traffic Engineering
No full textPart 1: The 9th Workshop on “5G – Putting Intelligence to the Network Edge” (5G-PINE)International audienceThe emerging 6G use cases will pose new challenges for the mobile network User Plane (UP), requiring its rapid evolution in terms of flexibility and intelligent optimisation. To achieve the foremost, the exploitation of the Software-Defined Networking (SDN) concept is commonly considered due to the logically centralised network control and native support for Traffic Engineering (TE). A promising solution to embed intelligence in the network is using Deep Reinforcement Learning (DRL) methods, which are capable of flexible optimisation of complex environments without prior modelling. While there exist several state of the art concepts combining the above technologies pair-wise, there is no approach that integrates them into a unified 6G-ready solution. This paper presents the novel 3GPP-compliant SDN-based UP architecture enhanced by DRL-based TE to facilitate emerging 6th Generation (6G) use cases. The approach leverages hierarchical architectures to improve the scalability of operations, support decentralised 6G network deployments and enable DRL usage in carrier-grade mobile networks
Resident-Oriented Green Energy Optimization Using a Multi-objective Evolutionary Algorithm
No full textPart 3: The 1st Workshop on “AI Applications for Achieving the Green Deal Targets” (ΑΙ4GD)International audienceThe European Green Deal has set ambitious short-term targets for reducing CO2 emissions and achieving climate neutrality. In communal living spaces, the associated challenges involve the exploitation of energy from renewable sources in order to reduce indirect CO2 emissions caused by grid electricity consumption, and the satisfaction of the residents, with their individual appliance-scheduling preferences that often conflict with their objective of minimizing associated billing charges. This paper tackles this multi-objective optimization problem by proposing a multi-objective evolutionary algorithm based on decomposition with decision making. The algorithm produces a set of optimal trade-offs between maximizing the satisfaction of resident appliance-scheduling preferences and minimizing their billing charges, with decision making opting for the trade-off offering minimal deviation from the use of green energy, consequently limiting the CO2 footprint. Our experimental evaluation, based on the energy consumption patterns of 10 UK households as recorded in the REFIT public dataset, demonstrates that the proposed approach clearly outperforms alternative state-of-the-art approaches
Digital Health, Development and Social Exclusion: DHIS2 and HIV Prevention Among Adolescent Girls, Young Women and Key Populations
No full textPart 3: Giving Voice to Marginalised Perspectives in IS ResearchInternational audienceThe dominant philosophy guiding digital health studies on interventions is that of ‘equity’ of healthcare services. This study takes on a different approach, that of equity of health, basing on a capability lens, which sees equity as a differential phenomenon, and extends this to technology development and implementation. This is particularly fitting in contexts where the focus is on populations who have traditionally been exposed to disparities in healthcare due to stigma and discrimination. The study follows an initiative which sought to promote uptake of health services by socially excluded members of society. The initiative was aimed at implementing digital health technologies to support adolescent girls, young women, and key populations in Zimbabwe. This study sought to understand the role digital health plays, and can play, in addressing the problems of social justice, and particularly health equity, in the context of socially deprived groups. It discusses the phenomenon in relation to emerging conceptualisations of health equity which go beyond utilitarian perspectives that focus on the distribution of healthcare services. The study does this by complementing conceptual ideas from Amartya Sen’s capability approach, with those from Shoshana Zuboff’s smart machine perspective. This helps to trace the specific ways in which technology is implicated in development and, specifically, social justice and equity
Unveiling the Smart Vision Emerging in ICT-Enabled Rural Development
No full textPart 8: Smart Collaborations and CrowdsourcingInternational audienceICT4D scholars note that broader visions of development generate serious consequences on IT-enabled development initiatives. However, much remains to deepen our current understanding about the interplay of emergent development visions and ICT4D initiatives in specific development sectors. This paper explores the vision of ‘smart rural development’ that is increasingly framing rural development goals and policies. Conceptual and empirical papers focusing on this issue, within the field of ICT4D, information systems, rural development studies and development studies, are reviewed. Findings from the review are interpreted with the lens of institutional logics perspective to develop a typology of three smart rural development visions. The typology offers alternative assumptions about rural transformation and ICT4D practices, thus opening new avenues for research and practice of ICT-enabled rural development