Early and Late Fusion for Multimodal Aggression Prediction in Dementia Patients: A Comparative Analysis

Applied Sciences, Journal Year: 2025, Volume and Issue: 15(11), P. 5823 - 5823

Published: May 22, 2025

Aggression in patients with dementia poses significant caregiving and clinical issues. In this work, fusion approaches—Early Fusion Late Fusion—were compared to classify aggression using audio visual signals. Early integrates the extracted features of two modalities into one dataset before classification, while prediction probabilities standalone classifiers a meta-classifier. Both models were tested Random Forest classifier five-fold cross-validation, performance was on accuracy, precision, recall, F1-score, ROC-AUC, inference time. The results showcase that is superior terms accuracy (0.876 vs. 0.828), recall (0.914 0.818), F1-score (0.867 0.835), ROC-AUC score (0.970 0.922), proving more suitable for high-sensitivity use cases like healthcare security. However, exhibited higher precision (0.852 0.824), indicating when false positives are requirement, preferable. Paired t-tests applied statistical comparison indicate alone significantly different, advantage Fusion. also performs slightly less time, which makes it real-time systems. These findings provide information multimodal strategies their applicability detection aggressive behavior, can contribute development efficient monitoring systems care.

Language: Английский

---

Research on 2D Animation Simulation Based on Artificial Intelligence and Biomechanical Modeling

EAI Endorsed Transactions on Pervasive Health and Technology, Journal Year: 2024, Volume and Issue: 10

Published: May 23, 2024

Animation techniques have been completely transformed by the union of Artificial Intelligence (AI) and biomechanical modeling, particularly in 2D animation. This study looks at a combination AI biomechanics to address challenges simulating Current approaches animation often struggle achieve lifelike fluid movements, especially when representing complex motion or interaction. These traditional rely on manual keyframing physics simulation, which may be time-consuming do not provide rich detail needed for realism animations. To meet these aspects, this suggested using with Biomechanical Modeling (2D-AI-BM). Our approach thus harnesses Deep Neural Network (DNN) moving forecasts improvement biopsychological principles help us imitate natural human actions better. In addition character animation, it could apply interactive storytelling educational simulations. As result, animators get more control over generation while drastically reducing necessity intervention through fusion biomechanics, smoothens production pipeline paper considers several important metrics evaluate proposed approach’s effectiveness, including user satisfaction, computational efficiency, smoothness realism. Comparative studies classical methods showed that method generates realistic movements characters saving time during production. The numerical findings exemplify recommended 2D-AI-BM model improves an accuracy rate 97.4%, efficiency ratio 96.3%, 95.4%, pose detection 94.8% scalability 93.2% compared other popular techniques.

Language: Английский

---

Innovative Applications of Virtual Reality in Cultural and Creative Product Design

Computer-Aided Design and Applications, Journal Year: 2024, Volume and Issue: unknown, P. 125 - 138

Published: June 1, 2024

Aiming at the problems of low rendering efficiency and long time-consuming traditional dynamic scene methods, this study proposes an optimization method animation design CAD virtual reality based on a transformer model.Through construction model, model is used for design.Combined with spatial data conversion graphics, geometric transformation carried out to realize graphics.By using primitive transformation, normal lighting models combined characteristics rendering, constructed.Experimental results show that, compared proposed can shorten time scenes effectively improve scenes.

Language: Английский

---

Advanced Yoga Pose Estimation: Enhancing PoseNet with Adaptive Key Point Elimination

Published: April 17, 2024

Yoga is an ancient practice famous for its physical and mental benefits assistance, gaining recognition as a means of achieving enhanced flexibility, strength, clarity. With the increasing demand accessible effective yoga instructions through online mode, in this work pose estimation model will be developed which may used with web application, to pro-vide lessons convenience one's own environment. This paper delves into integration advanced computer vision machine learning techniques, particularly PoseNet, accurately tracks human poses using 17 key points. But all points identified posenet architecture not useful estimating accuracy might result increased redundancy thus decreasing performance. To overcome drawback, innovative approach called adaptive point elimination technique proposed improve precision estimation. The methodology involves leveraging transfer fine-tune PoseNet specific employing custom Convolutional Neural Network (CNN) classification various postures. experimental methods are tested on dataset taken from Kaggle website following asanas: Chair, Cobra, Dog, Tree, Warrior poses. able achieve validation 98.678% final evaluation 99.32% test our system.

Language: Английский

---

Object detection in optical imaging of the Internet of Things based on deep learning

PeerJ Computer Science, Journal Year: 2023, Volume and Issue: 9, P. e1718 - e1718

Published: Dec. 11, 2023

This article endeavors to enhance image recognition technology within the context of Internet Things (IoT). A dynamic target detection training model is established through convolutional neural network (CNN) algorithm framework deep learning (DL). Three distinct configurations are proposed: a nine-layer convolution model, seven-layer and residual module model. Subsequently, simulation CNN based on optical imaging constructed, experiments conducted in scenarios simple salient environments, complex intricate micro-environment. By determining optimal iterations, comparisons drawn terms precision, accuracy, Intersection Over Union (IoU), frames per second (FPS) among different configurations. Finally, an attention mechanism incorporated DL framework, leading construction that operates at three difficulty levels: simple, intermediate, challenging. Through comparative analysis against prevalent algorithms, this delves into accuracy efficiency various models for IoT detection. Key findings include: (1) The exhibits commendable confidence although it encounters certain instances undetected images, indicating scope improvement. (2) when employing loss function comprising both mean square error (MSE) cross entropy, demonstrates superior performance manifesting high IoU, metrics, thereby establishing itself as robust (3) Within micro-environments, utilizing functions MSE yields substantial results, with FPS values amounting 0.99, 0.83, 29.9, respectively. (4) enriched outperforms other detection, achieving highest rates 24.86%, 17.8%, 14.77% challenging levels, Although entails slightly longer times, its overall excellent, augmenting effectiveness object IoT. strives speed, bolster capability systems, refine settings. implications encompass reduced manual costs provision theoretical foundation optimizing technologies context.

Language: Английский

---

Interactive Landscape Design Based on the Integration of Computer-Aided Design and Virtual Reality

Computer-Aided Design and Applications, Journal Year: 2024, Volume and Issue: unknown, P. 153 - 167

Published: June 1, 2024

Aiming at the problems of low rendering efficiency and long time-consuming traditional dynamic scene methods, this study proposes an optimization method animation design CAD virtual reality based on a transformer model.Through construction model, model is used for design.Combined with spatial data conversion graphics, geometric transformation carried out to realize graphics.By using primitive transformation, normal lighting models combined characteristics rendering, constructed.Experimental results show that, compared proposed can shorten time scenes effectively improve scenes.

Language: Английский

---

Graph Neural Networks-Based Virtual Reality Data Fusion Display for New Media Art

Computer-Aided Design and Applications, Journal Year: 2024, Volume and Issue: unknown, P. 97 - 109

Published: June 1, 2024

Aiming at the problems of low rendering efficiency and long time-consuming traditional dynamic scene methods, this study proposes an optimization method animation design CAD virtual reality based on a transformer model.Through construction model, model is used for design.Combined with spatial data conversion graphics, geometric transformation carried out to realize graphics.By using primitive transformation, normal lighting models combined characteristics rendering, constructed.Experimental results show that, compared proposed can shorten time scenes effectively improve scenes.

Language: Английский

---

Dynamic Display of Ceramic Model Based on Virtual Reality Environment

Computer-Aided Design and Applications, Journal Year: 2024, Volume and Issue: unknown, P. 69 - 82

Published: June 1, 2024

Aiming at the problems of low rendering efficiency and long time-consuming traditional dynamic scene methods, this study proposes an optimization method animation design CAD virtual reality based on a transformer model.Through construction model, model is used for design.Combined with spatial data conversion graphics, geometric transformation carried out to realize graphics.By using primitive transformation, normal lighting models combined characteristics rendering, constructed.Experimental results show that, compared proposed can shorten time scenes effectively improve scenes.

Language: Английский

---

Optimization of Advertising Design Model Driven by Deep Learning and Interactive VR Display

Computer-Aided Design and Applications, Journal Year: 2024, Volume and Issue: unknown, P. 182 - 195

Published: June 1, 2024

Aiming at the problems of low rendering efficiency and long time-consuming traditional dynamic scene methods, this study proposes an optimization method animation design CAD virtual reality based on a transformer model.Through construction model, model is used for design.Combined with spatial data conversion graphics, geometric transformation carried out to realize graphics.By using primitive transformation, normal lighting models combined characteristics rendering, constructed.Experimental results show that, compared proposed can shorten time scenes effectively improve scenes.

Language: Английский

---

Application of Graph Neural Network and Virtual Reality Based on the Concept of Sustainable Design

Computer-Aided Design and Applications, Journal Year: 2024, Volume and Issue: unknown, P. 15 - 27

Published: June 1, 2024

Aiming at the problems of low rendering efficiency and long time-consuming traditional dynamic scene methods, this study proposes an optimization method animation design CAD virtual reality based on a transformer model.Through construction model, model is used for design.Combined with spatial data conversion graphics, geometric transformation carried out to realize graphics.By using primitive transformation, normal lighting models combined characteristics rendering, constructed.Experimental results show that, compared proposed can shorten time scenes effectively improve scenes.

Language: Английский

---