KMITL

Permanent URI for this communityhttps://dspace.kmitl.ac.th/handle/123456789/1

Browse

Search Results

Now showing 1 - 10 of 12

Some of the metrics are blocked by your
consent settings
Multi-agent deep learning on tensor fields for segmentation of ultrasound images
(2026-03-01)
Sharma, Suman
;
Moodleah, Samart
;
Makhanov, Stanislav S.
Medical image analysis often relies on vector fields (VF), which are fundamental to deterministic models such as Active Contours, Level Set Methods, Phase Portrait Analysis, and artificial agent–based formulations. We experimentally demonstrate that a Deep Learning Neural Network (DLNN) capable of interpreting VF structures can substantially enhance the decision-making capabilities of artificial agents. We introduce a novel hybrid framework that integrates artificial life (AL) agents operating within a VF with a DLNN that guides their behavior. A key innovation of the model is the initialization of AL agents using streamlines derived from the VF orthogonal to the generalized gradient vector flow (GGVF) field. The VF is further transformed into a bi-directional Tensor Field (TF), where the spatial distribution and classification of degenerate points (DPs) serve as critical features. These DPs are leveraged to train AL agents through the DLNN, enabling them to follow meaningful anatomical structures. The framework employs DeepLabV3+ with ResNet50 as the backbone and is trained on 179 benign and 107 malignant breast ultrasound images collected at Thammasat University Hospital (TUH) and annotated by three leading radiologists, in addition to the BUSI and UDIAT datasets. Using 10-fold cross-validation, the proposed method achieves stable and robust performance across three datasets. Mean Dice scores of 94.84±1.63% (TUH), 94.16±1.62% (BUSI), and 93.67±1.51% (UDIAT) are obtained, with corresponding IoU values of 91.19±1.76%, 90.21±1.83% and 89.08±1.70%, demonstrating strong generalization across diverse imaging conditions. Comparative evaluations against state-of-the-art methods confirm the superiority of the proposed model. A video demonstration is available at: https://tinyurl.com/AL-DLNN.
Some of the metrics are blocked by your
consent settings
Automated Classification of User Exercise Poses in Virtual Reality Using Machine Learning-Based Human Pose Estimation
(2025-01-01)
Kongchansawang, Panuwat
;
Naowavathong, Thunchanok
;
Srisuttee, Ratakorn
;
Moodleah, Samart
Human Pose Estimation (HPE) using machine learning presents significant potential for objectively analyzing user movements within Virtual Reality (VR) environments, particularly for applications involving guided physical exercises or interactive tasks. Analyzing 2D video footage from VR sessions for HPE offers a computationally efficient approach for tracking user movements. However, accuracy can be compromised by various factors, including occlusions from clothing or the VR headset itself. This study develops and evaluates a robust HPE system specifically designed to accurately classify predefined poses performed by users within such VR environments. The proposed system utilizes MediaPipe for corporal landmark extraction from user images (for training) and employs a Stacking Classifier ensemble, with XGBClassifier as the meta-learner, to classify six key exercise poses. Key results demonstrate an overall pose classification accuracy of 94% on a dedicated test set, with certain poses like ‘Butterfly Hug’ and ‘Arm’ achieving 100% accuracy. These findings highlight the system’s potential to provide reliable, quantitative assessments of user adherence to prescribed movements in immersive interactive environments, offering valuable data for applications requiring objective analysis of physical performance in VR.
Some of the metrics are blocked by your
consent settings
AI-Powered MetaHuman Interviewer: Serious Game for Student Job Interview Skills
(2025-01-01)
Mongkoljaturong, Kansada
;
Manitsakulwong, Mesa
;
Moodleah, Samart
Effective interview preparation is critical for university graduates entering a competitive job market, yet traditional methods often lack sufficient realism and interactivity. This paper presents the development and evaluation of an innovative serious game, the Interactive Interview Assistant Application (IIAA), designed to address this gap. The IIAA simulates job interviews using an AI-driven digital human interviewer powered by MetaHuman technology, a Large Language Model (LLM) for dynamic dialogue, and integrated Speech-to-Text, Text-to-Speech, and Speech-to-Animation systems, offering a personalized and immersive practice environment. The system was evaluated with senior undergraduate students across three distinct career paths (UX/UI Design, Software Engineering, Game Development). The evaluation focused on system performance, overall usability, and qualitative user feedback. Results indicate that the IIAA successfully creates an engaging and realistic simulation, with users reporting high satisfaction regarding its ability to emulate real-world interview scenarios and provide valuable, actionable feedback. While identified performance latencies related to high-definition 3D animation and voice processing require further optimization, the study underscores the significant potential of this approach. This research demonstrates that sophisticated game-based tools can substantially enhance interview preparedness, thereby improving student confidence and readiness for professional employment.
Some of the metrics are blocked by your
consent settings
Learning by Doing: The Impact of Virtual Reality Scenarios on First Aid Training Effectiveness
(2024-01-01)
Kawmong, Chanikarn
;
Kuljittwattana, Ittiporn
;
Moodleah, Samart
Limited accessibility, high costs, and potential lack of user engagement can hinder the effectiveness of traditional first-aid training methods. Thailand serves as a case study, where data from 2017-2021 reveals a concerning incidence of emergency illnesses (12,000 per 100,000 people) and a concerning mortality rate (179.6 per 100,000 people). This study investigates the potential of Virtual Reality (VR) technology as a novel and engaging approach to first-aid t raining, s pecifically de signed for Thai speakers. Participants engaged with a VR first aid training game featuring interactive modules on essential equipment, emergency procedures, and Cardiopulmonary Resuscitation (CPR) with Automated External Defibrillator (AED) u sage. Compared to a control group, participants who trained with the VR First Aid game demonstrated a significant increase in first-aid knowledge (31.00%). Additionally, the VR training method yielded a substantial increase in user satisfaction (42.42%) compared to traditional approaches. These findings highlight V R technology as a promising alternative for first-aid training, potentially offering improved accessibility, engagement, and learning outcomes compared to traditional methods.
Some of the metrics are blocked by your
consent settings
Human Posture Detection in Virtual Reality Applications for Stress Reduction
(2024-01-01)
Kongchansawang, Panuwat
;
Wongsiripa, Santakorn
;
Moodleah, Samart
This research explores the potential of human pose estimation (HPE) using machine learning to analyze user movements within a virtual reality (VR) environment for stress assessment. By analyzing 2D video footage captured during VR sessions, HPE provides a computationally efficient method for tracking user poses, making it an ideal tool for evaluating therapeutic exercises. However, challenges such as clothing occlusions, background complexity, and varying lighting conditions can affect HPE accuracy. This study investigates the effectiveness of HPE in overcoming these challenges within a VR therapy context. We propose a robust system designed to accurately assess user posture during VR therapy simulations. Our approach leverages advanced machine learning algorithms to enhance the precision of pose estimation, even under challenging conditions. Preliminary results indicate that our system achieves an impressive overall accuracy of 94%, demonstrating its potential to provide reliable assessments of user movements.
Some of the metrics are blocked by your
consent settings
Edge-Driven Multi-Agent Reinforcement Learning: A Novel Approach to Ultrasound Breast Tumor Segmentation
(2023-12-01)
Karunanayake, Nalan
;
Moodleah, Samart
;
Makhanov, Stanislav S.
A segmentation model of the ultrasound (US) images of breast tumors based on virtual agents trained using reinforcement learning (RL) is proposed. The agents, living in the edge map, are able to avoid false boundaries, connect broken parts, and finally, accurately delineate the contour of the tumor. The agents move similarly to robots navigating in the unknown environment with the goal of maximizing the rewards. The individual agent does not know the goal of the entire population. However, since the robots communicate, the model is able to understand the global information and fit the irregular boundaries of complicated objects. Combining the RL with a neural network makes it possible to automatically learn and select the local features. In particular, the agents handle the edge leaks and artifacts typical for the US images. The proposed model outperforms 13 state-of-the-art algorithms, including selected deep learning models and their modifications.
Some of the metrics are blocked by your
consent settings
Adaptive Slicing of Point Cloud Directly with Discrete Interpolable-Area Error Profile in Additive Manufacturing
(2023-02-01)
Moodleah, Samart
;
Kirimasthong, Khwunta
Point cloud objects have gained popularity in three-dimensional (3D) printing recently due to advancements in reverse engineering technology. Fabricating an object with a fused deposition modeling (FDM) printer requires converting the object to layered contours, which involves a slicing process. The slicing process of a point cloud object usually requires reconstructing a 3D object from a point cloud, which requires users' deep understanding of 3D modeling software and a laborious work process. To avoid these problems, the direct slicing of point cloud objects is gaining more popularity. This research work proposes an adaptive slicing approach from point cloud objects directly without surface reconstruction. The adaptive slicing maintains the global geometry error while requiring a smaller number of fabrication layers and printing time. A new error profile used in the adaptive slicing approach is introduced. It approximates the geometry error from the point cloud directly based on the discrete interpolable-area (DIA) error between two adjacent layers. The interpolable capability of the DIA error profile allows the adaptive slicing algorithm to efficiently measure the geometry error of a point cloud. We perform the proposed algorithm with four point cloud models that represent both symmetrical and asymmetrical shapes. The adaptive slicing results show that the performance is increased by 8.05%-32.73% while maintaining accuracy compared to traditional uniform slicing. Furthermore, the fabrication time and materials used are reduced by 10.30%-39.10% and 1.01%-13.47%, respectively. Based on these results, further research can be focused on finding an optimal threshold between the accuracy of the contour projection and the distance between the layers, which could further improve fabrication performance.
Some of the metrics are blocked by your
consent settings
Naval Wargame Prototyping: Multiplayer Real-Time Strategy Game Simulation Using Unreal Engine
(2023-01-01)
Chavanit, Nattawat
;
Bualoy, Sukawit
;
Moodleah, Samart
We propose an integration framework for a wargame prototype using the modern game engine - Unreal Engine. This wargame functions as a simulation tool for strategy training, strategy testing, and simulating enemy forces like warships, aircraft, and weapons. Existing wargames come in proprietary and free versions. The former is often expensive and exclusive due to security reasons, while developing a functional wargame is complex, requiring various technological components such as a physics system and artificial intelligence (AI). To overcome these challenges, we propose a rapid prototype using Unreal Engine. This approach leverages advanced technology and ensures the prototype is ready for future upgrades when new versions of the game engine are released. We evaluate the prototype's system capabilities and expert assessments.
Some of the metrics are blocked by your
consent settings
A Development of Game-Based Learning in Virtual Reality for Fire Safety Training in Thailand
(2021-06-30)
Satapanasatien, Kodchaporn
;
Phuawiriyakul, Thanchanok
;
Moodleah, Samart
Fire incidents damaged both the economy and human life. In the past three decades, 59,387 fire incidents in Thailand approximately lost 40 billion baht and 2,076 deaths. Fire safety training methods are organized in a very limited number each year for many reasons such as training place or safety concern. We propose game-based learning for fire safety training using virtual reality technology. We create five learning lessons based on fire training contents and three playing stages (play, learn and test) that players can interact with. It behaves as a self-learning tool that can be used often and overcome the difficulty in organizing fire safety training. In addition, virtual technology simulates realistic computer graphic contents and rich interactive actions. Most importantly it offers a safe virtual environment; it extends the audience to a wide range of ages. Our game-based training is evaluated, and the result shows that fire safety knowledge is increased by 122% and 63% compared to the non-training and the traditional training respectively as well as users' satisfaction average score is exceeded 90%.
Some of the metrics are blocked by your
consent settings
Points using Localized Distance for Contour Generation from Point Cloud for 3D Printing
(2021-01-01)
Moodleah, Samart
;
Kirimasthong, Khwunta
We present a robust and simple method to select the most correlated points (in each layer) for layered contour generation from a 3D point cloud model in additive manufacturing. The contour projection of each layer point uses a planar least square projection technique. One critical step in contour projection is that the correlation between each projection point and other points (weights) on a slicing plane directly affects to the contour generation accuracy. The constructing contour from point cloud directly is a challenging task because there is no information of mesh topology, hence, no sequential order of points for contour generation. A search algorithm to find the most correlated points from skeletal points (reference) using the localized distance function is implemented. The experiment results show that our method reduces an average accuracy error for both wide and narrow point distributions by 15.57% to 35.20% and 4.28% to 12.78% respectively compared to the existina method.

KMITL

Browse

Filters

Settings

Sort By

Results per page

Search Results