Fadhlan Hafizhelmi Kamaru Zaman
Current Organizations / Institutes
The organizations/institutes from the scholar's latest article
Previous Organizations / Institutes
The organizations / institutes from the scholar's other articles in the past ten years
Research Subjects
Fadhlan Hafizhelmi Kamaru Zaman has a total of 86 co-authors. Here are the top co-authors by number of shared publications:
- Rohilah Sahak
Universiti Teknologi MARA
7 shared publications Syahrul Afzal Che Abdullah
Universiti Teknologi MARA, Malaysia
7 shared publications- Husna Zainol Abidin
Universiti Teknologi MARA
6 shared publications Ahmad Ihsan Mohd Yassin
Universiti Teknologi MARA
5 shared publications- Kok Mun Ng
Universiti Teknologi MARA
4 shared publications Nooritawati Md Tahir
Institute for Big Data Analytics and Artificial Intelligence (IBDAAI), Universiti Teknologi Mara (UiTM), Malaysia
4 shared publications
Latest Publications
- Top Cited PapersOpen Access
- 2 January 2026
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Open Journal of Intelligent Transportation Systems
- Vol. 7, 233-267
Transformer Architectures for Distracted Driving Behavior Detection: A Comprehensive Review of Vision-Based Approaches
Distracted driving is a leading cause of road accidents, with visual and manual distractions being particularly prevalent. Traditional computer vision methods, particularly Convolutional Neural Networks (CNNs), have been extensively utilized for detecting driver behavior; however, they face challenges in effectively modeling long-range dependencies and complex spatiotemporal patterns. Recent advancements in Vision Transformer (ViT) demonstrate significant potential to address these limitations by leveraging global attention mechanisms and a scalable architecture. This review presents a comprehensive review of ViT-based approaches in distracted driving detection, which covers both image-based and video-based methods. It examines several architectural innovations, such as lightweight ViT variants, pose-aware attention-enhanced models, and hybrid ViT-architecture designs. The review also explores multi-modal and multi-view fusion strategies, which integrate several inputs such as RGB, infrared, depth, and physiological signals to enhance model robustness across diverse driving scenarios. In addition, the paper highlights benchmark datasets and performance comparisons used in distracted driving behavior detection. Finally, this review highlights the current challenges, including computational cost and interpretability, while also proposing directions for future research. Overall, ViT-based models present a promising foundation for developing the next generation of intelligent driver monitoring systems.
- 2 January 2026
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Open Journal of Intelligent Transportation Systems
- Vol. 7, 233-267
Top Cited PapersOpen AccessTransformer Architectures for Distracted Driving Behavior Detection: A Comprehensive Review of Vision-Based Approaches
Distracted driving is a leading cause of road accidents, with visual and manual distractions being particularly prevalent. Traditional computer vision methods, particularly Convolutional Neural Networks (CNNs), have been extensively utilized for detecting driver behavior; however, they face challenges in effectively modeling long-range dependencies and complex spatiotemporal patterns. Recent advancements in Vision Transformer (ViT) demonstrate significant potential to address these limitations by leveraging global attention mechanisms and a scalable architecture. This review presents a comprehensive review of ViT-based approaches in distracted driving detection, which covers both image-based and video-based methods. It examines several architectural innovations, such as lightweight ViT variants, pose-aware attention-enhanced models, and hybrid ViT-architecture designs. The review also explores multi-modal and multi-view fusion strategies, which integrate several inputs such as RGB, infrared, depth, and physiological signals to enhance model robustness across diverse driving scenarios. In addition, the paper highlights benchmark datasets and performance comparisons used in distracted driving behavior detection. Finally, this review highlights the current challenges, including computational cost and interpretability, while also proposing directions for future research. Overall, ViT-based models present a promising foundation for developing the next generation of intelligent driver monitoring systems.
- Open Access
- 15 May 2025
- Published by IIUM Press in IIUM Engineering Journal
- Vol. 26 (2) , 169-186
Comparative Analysis of Vision Transformers and CNN Models for Driver Fatigue Classification
This study provides a comprehensive evaluation of Convolutional Neural Network (CNN) and Vision Transformer (ViT) models for driver fatigue classification, a critical issue in road safety. Using a custom driving behavior dataset, state-of-the-art CNN and ViT architectures, including VGG16, EfficientNet, MobileNet, Inception, DenseNet, ResNet, ViT, and Swin Transformer, were analyzed in this study to determine the best model for practical driver fatigue monitoring systems. Performance metrics such as accuracy, F1-score, training time, inference time, and frames per second (fps) were assessed across different hardware platforms, including a high-performance workstation, Raspberry Pi 5, and a desktop with a Graphic Processing Unit (GPU). Results demonstrate that CNN models, particularly VGG16, achieve the best balance between accuracy and efficiency, with an F1-score of 0.97 and 77.00 fps on a desktop. On the other hand, Swin V2S outperforms all models in terms of accuracy, achieving an F1-score of 0.99 and 61.18 fps on a GPU, although it exhibits limited efficiency on embedded systems. This study significantly contributes by providing practical recommendations for selecting models based on performance needs and hardware constraints, highlighting the suitability of ViTs for high-computation environments. The findings support the development of more efficient driver fatigue monitoring systems, offering practical implications for enhancing road safety and reducing traffic accidents. ABSTRAK: Kajian ini merupakan penilaian komprehensif terhadap model Konvolusi Rangkaian Neural (CNN) dan Transformer Penglihatan (ViT) bagi pengelasan keletihan pemandu, iaitu satu isu kritikal dalam keselamatan jalan raya. Menggunakan set data tingkah laku pemanduan tersuai, seni bina terkini CNN dan ViT, termasuk VGG16, EfficientNet, MobileNet, Inception, DenseNet, ResNet, ViT dan Transformer Swin dianalisa dalam kajian ini bagi menentukan model terbaik bagi sistem pemantauan keletihan pemandu yang praktikal. Metrik prestasi seperti ketepatan, skor F1, masa latihan, masa inferens, dan bingkai sesaat (fps) telah dinilai merentasi pelbagai platfom perkakasan, termasuk stesen kerja berprestasi tinggi, Raspberry Pi 5, dan komputer meja dengan Unit Pemprosesan Grafik (GPU). Dapatan kajian menunjukkan bahawa model CNN, khususnya VGG16, mencapai keseimbangan terbaik antara ketepatan dan kecekapan, dengan skor F1 sebanyak 0.97 dan 77.00 fps pada komputer meja. Sebaliknya, Swin V2S mengatasi semua model dari segi ketepatan, mencapai skor F1 sebanyak 0.99 dan 61.18 fps pada GPU, walaupun menunjukkan kecekapan yang terhad pada sistem terbenam. Kajian ini memberikan sumbangan yang signifikan dengan menyediakan cadangan praktikal bagi pemilihan model berdasarkan keperluan prestasi dan kekangan perkakasan, serta menonjolkan kesesuaian ViT bagi persekitaran berkomputasi tinggi. Penemuan ini menyokong pembangunan sistem pemantauan keletihan pemandu yang lebih cekap, dengan implikasi praktikal bagi meningkatkan keselamatan jalan raya dan mengurangkan kemalangan.
- 15 May 2025
- Published by IIUM Press in IIUM Engineering Journal
- Vol. 26 (2) , 169-186
Open AccessComparative Analysis of Vision Transformers and CNN Models for Driver Fatigue Classification
This study provides a comprehensive evaluation of Convolutional Neural Network (CNN) and Vision Transformer (ViT) models for driver fatigue classification, a critical issue in road safety. Using a custom driving behavior dataset, state-of-the-art CNN and ViT architectures, including VGG16, EfficientNet, MobileNet, Inception, DenseNet, ResNet, ViT, and Swin Transformer, were analyzed in this study to determine the best model for practical driver fatigue monitoring systems. Performance metrics such as accuracy, F1-score, training time, inference time, and frames per second (fps) were assessed across different hardware platforms, including a high-performance workstation, Raspberry Pi 5, and a desktop with a Graphic Processing Unit (GPU). Results demonstrate that CNN models, particularly VGG16, achieve the best balance between accuracy and efficiency, with an F1-score of 0.97 and 77.00 fps on a desktop. On the other hand, Swin V2S outperforms all models in terms of accuracy, achieving an F1-score of 0.99 and 61.18 fps on a GPU, although it exhibits limited efficiency on embedded systems. This study significantly contributes by providing practical recommendations for selecting models based on performance needs and hardware constraints, highlighting the suitability of ViTs for high-computation environments. The findings support the development of more efficient driver fatigue monitoring systems, offering practical implications for enhancing road safety and reducing traffic accidents. ABSTRAK: Kajian ini merupakan penilaian komprehensif terhadap model Konvolusi Rangkaian Neural (CNN) dan Transformer Penglihatan (ViT) bagi pengelasan keletihan pemandu, iaitu satu isu kritikal dalam keselamatan jalan raya. Menggunakan set data tingkah laku pemanduan tersuai, seni bina terkini CNN dan ViT, termasuk VGG16, EfficientNet, MobileNet, Inception, DenseNet, ResNet, ViT dan Transformer Swin dianalisa dalam kajian ini bagi menentukan model terbaik bagi sistem pemantauan keletihan pemandu yang praktikal. Metrik prestasi seperti ketepatan, skor F1, masa latihan, masa inferens, dan bingkai sesaat (fps) telah dinilai merentasi pelbagai platfom perkakasan, termasuk stesen kerja berprestasi tinggi, Raspberry Pi 5, dan komputer meja dengan Unit Pemprosesan Grafik (GPU). Dapatan kajian menunjukkan bahawa model CNN, khususnya VGG16, mencapai keseimbangan terbaik antara ketepatan dan kecekapan, dengan skor F1 sebanyak 0.97 dan 77.00 fps pada komputer meja. Sebaliknya, Swin V2S mengatasi semua model dari segi ketepatan, mencapai skor F1 sebanyak 0.99 dan 61.18 fps pada GPU, walaupun menunjukkan kecekapan yang terhad pada sistem terbenam. Kajian ini memberikan sumbangan yang signifikan dengan menyediakan cadangan praktikal bagi pemilihan model berdasarkan keperluan prestasi dan kekangan perkakasan, serta menonjolkan kesesuaian ViT bagi persekitaran berkomputasi tinggi. Penemuan ini menyokong pembangunan sistem pemantauan keletihan pemandu yang lebih cekap, dengan implikasi praktikal bagi meningkatkan keselamatan jalan raya dan mengurangkan kemalangan.
- Open Access
- 1 April 2025
- Published by AIP Publishing in Journal of Applied Physics
- Vol. 137 (13)
Diffusion models for topology optimization in 3D printing applications
Structural topology optimization, a critical aspect of engineering design in aerospace, mechanical, and civil engineering, seeks to discover the ideal physical structure for optimizing mechanical performance, particularly, in 3D structures. In recent times, Generative Adversarial Networks (GANs) and autoencoders have gained popularity as an alternative to traditional iterative topology optimization methods. However, these models often pose challenges in terms of training, exhibit limited generalizability, and prioritize mimicking optimal structures while neglecting factors such as manufacturability and mechanical compliance. In response to these limitations, we introduce an architecture based on conditional diffusion models aimed at achieving topology optimization that is not only performance-aware but also manufacturability-aware, specifically for 3D structures. Our approach incorporates a surrogate model-based guidance strategy that actively promotes structures with low compliance and excellent manufacturability. The results of our method surpass those of a state-of-the-art conditional GAN and conditional variational autoencoder (VAE), reducing the average error in physical performance by a factor of eight and generating eleven times fewer infeasible samples. By introducing diffusion models to 3D topology optimization, we demonstrate the superior capabilities of conditional diffusion models in engineering design synthesis applications as well. Furthermore, our work offers a broader framework for tackling engineering optimization problems using diffusion models and external performance, all while considering constraint-aware guidance.
- 1 April 2025
- Published by AIP Publishing in Journal of Applied Physics
- Vol. 137 (13)
Open AccessDiffusion models for topology optimization in 3D printing applications
Structural topology optimization, a critical aspect of engineering design in aerospace, mechanical, and civil engineering, seeks to discover the ideal physical structure for optimizing mechanical performance, particularly, in 3D structures. In recent times, Generative Adversarial Networks (GANs) and autoencoders have gained popularity as an alternative to traditional iterative topology optimization methods. However, these models often pose challenges in terms of training, exhibit limited generalizability, and prioritize mimicking optimal structures while neglecting factors such as manufacturability and mechanical compliance. In response to these limitations, we introduce an architecture based on conditional diffusion models aimed at achieving topology optimization that is not only performance-aware but also manufacturability-aware, specifically for 3D structures. Our approach incorporates a surrogate model-based guidance strategy that actively promotes structures with low compliance and excellent manufacturability. The results of our method surpass those of a state-of-the-art conditional GAN and conditional variational autoencoder (VAE), reducing the average error in physical performance by a factor of eight and generating eleven times fewer infeasible samples. By introducing diffusion models to 3D topology optimization, we demonstrate the superior capabilities of conditional diffusion models in engineering design synthesis applications as well. Furthermore, our work offers a broader framework for tackling engineering optimization problems using diffusion models and external performance, all while considering constraint-aware guidance.
- Open Access
- 21 November 2024
- Published by Universitas Muhammadiyah Magelang in Automotive Experiences
- Vol. 7 (3) , 513-524
Malaysian Public”™s Perception Toward Event Data Recorder (EDR) in Vehicles
The Event Data Recorder (EDR) is an important device in a vehicle that can be used to analyze vehicle accidents. EDRs record and store crucial sensor data before, during, and after accidents, which can be used in reconstructing accident events. EDR has been regulated in the US and UK; however, its implementation in the ASEAN region, especially in Malaysia, is relatively new. In this study, a survey is conducted to investigate the perception of Malaysians toward EDR. There are three focuses of this survey: (1) the public's awareness of EDR's existence; (2) their perception of the benefits of EDR; and (3) their privacy concerns with the use of EDR in vehicles. The survey revealed that a majority (75.4%) of respondents were unaware of the existence of EDR, but 72.3% acknowledged that EDR could help identify accident causes, and 59.5% believed it could promote safer driving. Moreover, over 66% of respondents agreed that EDR could enhance vehicle and road safety. Besides, 40.3% expressed concern about potential privacy breaches and misuse of EDR data. Despite that, nearly 80% of respondents were in favor of installing EDR in their vehicles and allowing the data to be used in court. Additionally, 70% indicated that EDR would become a criterion for vehicle selection, and they would support mandatory EDR regulations.
- 21 November 2024
- Published by Universitas Muhammadiyah Magelang in Automotive Experiences
- Vol. 7 (3) , 513-524
Open AccessMalaysian Public”™s Perception Toward Event Data Recorder (EDR) in Vehicles
The Event Data Recorder (EDR) is an important device in a vehicle that can be used to analyze vehicle accidents. EDRs record and store crucial sensor data before, during, and after accidents, which can be used in reconstructing accident events. EDR has been regulated in the US and UK; however, its implementation in the ASEAN region, especially in Malaysia, is relatively new. In this study, a survey is conducted to investigate the perception of Malaysians toward EDR. There are three focuses of this survey: (1) the public's awareness of EDR's existence; (2) their perception of the benefits of EDR; and (3) their privacy concerns with the use of EDR in vehicles. The survey revealed that a majority (75.4%) of respondents were unaware of the existence of EDR, but 72.3% acknowledged that EDR could help identify accident causes, and 59.5% believed it could promote safer driving. Moreover, over 66% of respondents agreed that EDR could enhance vehicle and road safety. Besides, 40.3% expressed concern about potential privacy breaches and misuse of EDR data. Despite that, nearly 80% of respondents were in favor of installing EDR in their vehicles and allowing the data to be used in court. Additionally, 70% indicated that EDR would become a criterion for vehicle selection, and they would support mandatory EDR regulations.
- Open Access
- 22 October 2024
- Published by Akademia Baru Publishing in CFD Letters
- Vol. 17 (3) , 109-123
Parametric Optimization For Power Generation Of Flow Induced Vibration Energy Harvester
Flow-induced vibration occurs when the motion of fluids through a structure induces oscillations or vibrations in the structure. An effective flow-induced vibration energy harvester has substantial challenges due to the river's irregular velocity flows. It is not practicable to use one parameter for all velocities. This work presents the testing of a flow-induced vibrational energy harvester in laminar flow using two circular cylinders positioned in tandem within an open-channel flow. A CFD simulation using COMSOL Multiphysics was performed for the proposed parameter. A comprehensive simulation run at multiple Reynolds numbers with varying gap lengths between the bluff bodies is studied to determine the maximum power generated. Simulation results show that the optimal gap lengths for Re 60, 80, 100, 120, 140, and 160 are 8.5, 6.0, 3.0, 3.0, 3.5, and 4.5, respectively. These gap lengths result in power outputs of 0.0315 W, 2.616 W, 1.899 W, 0.6552 W, 0.5018 W, and 0.3782 W. By demonstrating the relationship between Reynolds number and gap length, this study provides important information for maximising the energy harvesting from flow-induced vibration (FIV).
- 22 October 2024
- Published by Akademia Baru Publishing in CFD Letters
- Vol. 17 (3) , 109-123
Open AccessParametric Optimization For Power Generation Of Flow Induced Vibration Energy Harvester
Flow-induced vibration occurs when the motion of fluids through a structure induces oscillations or vibrations in the structure. An effective flow-induced vibration energy harvester has substantial challenges due to the river's irregular velocity flows. It is not practicable to use one parameter for all velocities. This work presents the testing of a flow-induced vibrational energy harvester in laminar flow using two circular cylinders positioned in tandem within an open-channel flow. A CFD simulation using COMSOL Multiphysics was performed for the proposed parameter. A comprehensive simulation run at multiple Reynolds numbers with varying gap lengths between the bluff bodies is studied to determine the maximum power generated. Simulation results show that the optimal gap lengths for Re 60, 80, 100, 120, 140, and 160 are 8.5, 6.0, 3.0, 3.0, 3.5, and 4.5, respectively. These gap lengths result in power outputs of 0.0315 W, 2.616 W, 1.899 W, 0.6552 W, 0.5018 W, and 0.3782 W. By demonstrating the relationship between Reynolds number and gap length, this study provides important information for maximising the energy harvesting from flow-induced vibration (FIV).
- Open Access
- 14 July 2024
- Published by IIUM Press in IIUM Engineering Journal
- Vol. 25 (2) , 325-337
Exploring the Efficacy of Piezoelectric-Based Sensory Systems for Heart Rate Monitoring in Differentiating Stress vs. Relax Conditions
Stress has diverse effects on human physiological reactions, and one such effect is on heart rate (HR). The established methods to acquire HR is by electrocardiogram (ECG) and photoplethysmogram (PPG). ECG electrodes need to be placed on the chest, which can cause inconvenience and is not practical in daily life, while PPG signals are known to contain more noise than ECG. Thus, this work aims to investigate the efficacy of a piezoelectric-based sensory system in measuring HR and using the signal to differentiate stressed and relaxed conditions by means of statistical analysis. Two activities were conducted to achieve the goal. The first experiment involved collecting and analysing piezoelectric signals to measure the pulse rate (bpm) and compare this with the HR from PPG. For the second experiment, the piezoelectric-based HR was calculated from 20 subjects (male and female, age ranging between 20 and 25) in relaxed and stressed conditions. The stress condition was triggered using two stressors: the Stroop Colour Word Test and the Digit Span Test. Statistical analyses reveal a strong positive correlation between piezoelectric-based heart rate (HR) and oximeter readings (r(12) = 0.993, p < 0.001), despite the fact that the values are not precisely identical. In addition, the findings also indicate that there are significant effects by the mental states (stressed and relaxed) on the piezoelectric-based HR readings (p<0.05). Employing a within-subject design condition, the results further illustrated that piezoelectric readings are elevated during stressed conditions (Mean±SD = 72.395±0.097) and diminished during relaxed conditions (Mean±SD = 71.615±0.126). Therefore, the suggested piezoelectric-based sensory system has been validated as an effective means of categorizing stress and relaxation based on heart rate signals. ABSTRAK: Tekanan mempunyai pelbagai kesan terhadap reaksi fisiologi manusia dan satu daripadanya adalah kadar denyut jantung (HR). Kaedah biasa bagi mengetahui HR adalah melalui elektrokardiogram (ECG) dan fotofetismogram (PPG). Elektrod ECG perlu dipasang pada dada di mana boleh menyebabkan ketidakselesaan dan tidak praktikal dalam kehidupan seharian, manakala isyarat PPG diketahui mengandungi lebih banyak bunyi berbanding ECG. Oleh itu, kajian ini bertujuan mengkaji kecekapan ystem deria berdasarkan piezoeletrik bagi mengukur HR dan menggunakan isyarat ini bagi membezakan keadaan tertekan atau tenang melalui cara analisis ystemic. Bagi mencapai tujuan ini, dua aktiviti dijalankan. Pertama eksperimen melibatkan pengumpulan dan analisis isyarat piezoeletrik bagi mengukur kadar nadi (bpm) dan membandingkan HR daripada PPG. Eksperimen kedua, HR berdasarkan piezoelektrik dikira dari 20 subjek (lelaki dan ystemic, berumur antara 20 dan 25) dalam keadaan tenang dan tertekan. Keadaan tertekan dibuat melalui dua ystemi: Ujian Stroop Patah Warna dan Ujian Rentang Digit. Analisis ystemic mendedahkan hubungkait yang kuat antara kadar denyut jantung (HR) berdasarkan piezoeletrik dan bacaan oksimeter (r(12) = 0.993, p < 0.001), walaupun bacaan tidak benar-benar serupa. Tambahan, penemuan ini menunjukkan terdapat kesan penting terhadap tahap mental (tertekan dan tenang) pada bacaan denyut jantung (HR) berdasarkan piezoeletrik (p<0.05). Dengan mengaplikasi keadaan reka bentuk subjek, dapatan menunjukkan bacaan piezoeletrik meningkat ystem keadaan tertekan (Mean±SD = 72.395±0.097) dan kurang ystem keadaan tenang (Mean±SD = 71.615±0.126). Oleh itu, ystem deria berdasarkan piezoeletrik telah diuji berkesan dalam mengkategori keadaan tertekan dan tenang berdasarkan isyarat denyut jantung.
- 14 July 2024
- Published by IIUM Press in IIUM Engineering Journal
- Vol. 25 (2) , 325-337
Open AccessExploring the Efficacy of Piezoelectric-Based Sensory Systems for Heart Rate Monitoring in Differentiating Stress vs. Relax Conditions
Stress has diverse effects on human physiological reactions, and one such effect is on heart rate (HR). The established methods to acquire HR is by electrocardiogram (ECG) and photoplethysmogram (PPG). ECG electrodes need to be placed on the chest, which can cause inconvenience and is not practical in daily life, while PPG signals are known to contain more noise than ECG. Thus, this work aims to investigate the efficacy of a piezoelectric-based sensory system in measuring HR and using the signal to differentiate stressed and relaxed conditions by means of statistical analysis. Two activities were conducted to achieve the goal. The first experiment involved collecting and analysing piezoelectric signals to measure the pulse rate (bpm) and compare this with the HR from PPG. For the second experiment, the piezoelectric-based HR was calculated from 20 subjects (male and female, age ranging between 20 and 25) in relaxed and stressed conditions. The stress condition was triggered using two stressors: the Stroop Colour Word Test and the Digit Span Test. Statistical analyses reveal a strong positive correlation between piezoelectric-based heart rate (HR) and oximeter readings (r(12) = 0.993, p < 0.001), despite the fact that the values are not precisely identical. In addition, the findings also indicate that there are significant effects by the mental states (stressed and relaxed) on the piezoelectric-based HR readings (p<0.05). Employing a within-subject design condition, the results further illustrated that piezoelectric readings are elevated during stressed conditions (Mean±SD = 72.395±0.097) and diminished during relaxed conditions (Mean±SD = 71.615±0.126). Therefore, the suggested piezoelectric-based sensory system has been validated as an effective means of categorizing stress and relaxation based on heart rate signals. ABSTRAK: Tekanan mempunyai pelbagai kesan terhadap reaksi fisiologi manusia dan satu daripadanya adalah kadar denyut jantung (HR). Kaedah biasa bagi mengetahui HR adalah melalui elektrokardiogram (ECG) dan fotofetismogram (PPG). Elektrod ECG perlu dipasang pada dada di mana boleh menyebabkan ketidakselesaan dan tidak praktikal dalam kehidupan seharian, manakala isyarat PPG diketahui mengandungi lebih banyak bunyi berbanding ECG. Oleh itu, kajian ini bertujuan mengkaji kecekapan ystem deria berdasarkan piezoeletrik bagi mengukur HR dan menggunakan isyarat ini bagi membezakan keadaan tertekan atau tenang melalui cara analisis ystemic. Bagi mencapai tujuan ini, dua aktiviti dijalankan. Pertama eksperimen melibatkan pengumpulan dan analisis isyarat piezoeletrik bagi mengukur kadar nadi (bpm) dan membandingkan HR daripada PPG. Eksperimen kedua, HR berdasarkan piezoelektrik dikira dari 20 subjek (lelaki dan ystemic, berumur antara 20 dan 25) dalam keadaan tenang dan tertekan. Keadaan tertekan dibuat melalui dua ystemi: Ujian Stroop Patah Warna dan Ujian Rentang Digit. Analisis ystemic mendedahkan hubungkait yang kuat antara kadar denyut jantung (HR) berdasarkan piezoeletrik dan bacaan oksimeter (r(12) = 0.993, p < 0.001), walaupun bacaan tidak benar-benar serupa. Tambahan, penemuan ini menunjukkan terdapat kesan penting terhadap tahap mental (tertekan dan tenang) pada bacaan denyut jantung (HR) berdasarkan piezoeletrik (p<0.05). Dengan mengaplikasi keadaan reka bentuk subjek, dapatan menunjukkan bacaan piezoeletrik meningkat ystem keadaan tertekan (Mean±SD = 72.395±0.097) dan kurang ystem keadaan tenang (Mean±SD = 71.615±0.126). Oleh itu, ystem deria berdasarkan piezoeletrik telah diuji berkesan dalam mengkategori keadaan tertekan dan tenang berdasarkan isyarat denyut jantung.
- Open Access
- 31 May 2024
- Published by Penerbit UTM Press in ASEAN Engineering Journal
- Vol. 14 (2) , 53-68
ENHANCEMENT OF TRAFFIC IMAGES UNDER DIFFERENT WEATHER CONDITIONS USING PYNET
Convolutional neural networks have real practical application potentials, such as for autonomous driving and semantic segmentation, but they are known to be sensitive to image degradations and corruption. Over the years, image enhancement in deep learning has shown drastic improvement. However, the clarity and quality of images still badly affect the robustness of semantic segmentation, especially for traffic images under different weather conditions. This paper proposes image enhancement performance using PyNET deep learning methods for more robust semantic segmentation of traffic images under different conditions. This work also proposes a new metric to objectively estimate the performance known as Image Similarity Metrics (ISM). Modification to PyNET is made in this work to allow this deep learning model to be used to enhance traffic images under various weather conditions such as fog, night, and rain. We compared PyNET performance against the Deep Convolutional Networks (DPED) and the Cycle Generative Adversarial Networks (CycleGAN) to evaluate the improvement gained by these image enhancement methods. Based on our experiment, PyNET gives the best image enhancement performance among those three methods in all weather conditions according to our proposed ISM. To support the validity of the ISM result, we performed tests by semantic segmentation of traffic images using ResNet-18 on the PyNET-enhanced images. Based on semantic segmentation results, PyNET improves the semantic segmentation of traffic images under different weather conditions by as much as 17% accuracy and delivers performance that directly validates the ISM scores, by showing that PyNET delivers the best semantic segmentation improvement in fog and night images
- 31 May 2024
- Published by Penerbit UTM Press in ASEAN Engineering Journal
- Vol. 14 (2) , 53-68
Open AccessENHANCEMENT OF TRAFFIC IMAGES UNDER DIFFERENT WEATHER CONDITIONS USING PYNET
Convolutional neural networks have real practical application potentials, such as for autonomous driving and semantic segmentation, but they are known to be sensitive to image degradations and corruption. Over the years, image enhancement in deep learning has shown drastic improvement. However, the clarity and quality of images still badly affect the robustness of semantic segmentation, especially for traffic images under different weather conditions. This paper proposes image enhancement performance using PyNET deep learning methods for more robust semantic segmentation of traffic images under different conditions. This work also proposes a new metric to objectively estimate the performance known as Image Similarity Metrics (ISM). Modification to PyNET is made in this work to allow this deep learning model to be used to enhance traffic images under various weather conditions such as fog, night, and rain. We compared PyNET performance against the Deep Convolutional Networks (DPED) and the Cycle Generative Adversarial Networks (CycleGAN) to evaluate the improvement gained by these image enhancement methods. Based on our experiment, PyNET gives the best image enhancement performance among those three methods in all weather conditions according to our proposed ISM. To support the validity of the ISM result, we performed tests by semantic segmentation of traffic images using ResNet-18 on the PyNET-enhanced images. Based on semantic segmentation results, PyNET improves the semantic segmentation of traffic images under different weather conditions by as much as 17% accuracy and delivers performance that directly validates the ISM scores, by showing that PyNET delivers the best semantic segmentation improvement in fog and night images
- Open Access
- 1 April 2024
- Published by Institute of Advanced Engineering and Science in Indonesian Journal of Electrical Engineering and Computer Science
- Vol. 34 (1) , 162-172
Modelling and estimating trajectory points from RTK-GNSS based on an integrated modelling approach
The sparse Gaussian process regression (GPR) has been used to model trajectory data from Real time kinematics-global navigation satellite system (RTK-GNSS). However, upon scrutinizing the model residuals; the sparse GPR model poorly fits the data and exhibits presence of correlated noise. This work attempts to address these issues by proposing an integrated modeling approach called GPR-LR-ARIMA where the sparse GPR was integrated with the linear regression with autoregressive integrated moving average errors (LR-ARIMA) to further enhance the description of the trajectory data. In this integrated approach, the predicted trajectory points from the GPR were further described by the LR-ARIMA. Simulation of the GPR-LR-ARIMA on three sets of trajectory data indicated better model fit, revealed in the normally distributed model residuals and symmetrically distributed scatter plots. Correlated noise was also successfully eliminated by the model. The GPR-LR-ARIMA outperformed both the GPR and LRARIMA by its ability to improve mean-absolute-error in 2-dimension positioning by up to 86%. The GPR-LR-ARIMA contributes to enhancement of positioning accuracy of dynamic GNSS measurements in localization and navigation system with good model fit.
- 1 April 2024
- Published by Institute of Advanced Engineering and Science in Indonesian Journal of Electrical Engineering and Computer Science
- Vol. 34 (1) , 162-172
Open AccessModelling and estimating trajectory points from RTK-GNSS based on an integrated modelling approach
The sparse Gaussian process regression (GPR) has been used to model trajectory data from Real time kinematics-global navigation satellite system (RTK-GNSS). However, upon scrutinizing the model residuals; the sparse GPR model poorly fits the data and exhibits presence of correlated noise. This work attempts to address these issues by proposing an integrated modeling approach called GPR-LR-ARIMA where the sparse GPR was integrated with the linear regression with autoregressive integrated moving average errors (LR-ARIMA) to further enhance the description of the trajectory data. In this integrated approach, the predicted trajectory points from the GPR were further described by the LR-ARIMA. Simulation of the GPR-LR-ARIMA on three sets of trajectory data indicated better model fit, revealed in the normally distributed model residuals and symmetrically distributed scatter plots. Correlated noise was also successfully eliminated by the model. The GPR-LR-ARIMA outperformed both the GPR and LRARIMA by its ability to improve mean-absolute-error in 2-dimension positioning by up to 86%. The GPR-LR-ARIMA contributes to enhancement of positioning accuracy of dynamic GNSS measurements in localization and navigation system with good model fit.
- Open Access
- 1 March 2024
- Published by Institute of Advanced Engineering and Science in Indonesian Journal of Electrical Engineering and Computer Science
- Vol. 33 (3) , 1974-1988
Improving night driving behavior recognition with ResNet50
The issue of driving behavior at night poses significant challenges due to reduced visibility and increased risk of accidents. Recent works have leveraged deep learning techniques to enhance night-time driving safety. However, the limited availability of high-quality training data and the lack of robustness in existing models present significant problems. In this work, we propose a novel approach to improve driving behavior recognition at night using ResNet50 with contrast limited adapted histogram equalization (CLAHE). We collected a new dataset and developed a more effective and robust model that can accurately recognize driving behaviors under low-illumination conditions, thereby reducing the likelihood of collisions and improving overall road safety. The experimental results demonstrate significant improvements in the deep learning model’s performance compared to conventional methods. Notably, the ResNet50 model delivers the best performance with accuracy rates of 90.73% using NIGHT-VIS-CLAHE data, demonstrating a 16% improvement in accuracy. For benchmark purposes, the InceptionV3, GoogleNet, and MobileNetV2 models also show enhanced accuracy through CLAHE implementation. Furthermore, NIGHT-VIS-CLAHE implementation in ResNet50 achieved 90.29% accuracy, surpassing the best NIGHT-IR InceptionV3 at 89.27%, highlighting the advantage of ResNet50 with CLAHE in low-light conditions even against infra-red sensor.
- 1 March 2024
- Published by Institute of Advanced Engineering and Science in Indonesian Journal of Electrical Engineering and Computer Science
- Vol. 33 (3) , 1974-1988
Open AccessImproving night driving behavior recognition with ResNet50
The issue of driving behavior at night poses significant challenges due to reduced visibility and increased risk of accidents. Recent works have leveraged deep learning techniques to enhance night-time driving safety. However, the limited availability of high-quality training data and the lack of robustness in existing models present significant problems. In this work, we propose a novel approach to improve driving behavior recognition at night using ResNet50 with contrast limited adapted histogram equalization (CLAHE). We collected a new dataset and developed a more effective and robust model that can accurately recognize driving behaviors under low-illumination conditions, thereby reducing the likelihood of collisions and improving overall road safety. The experimental results demonstrate significant improvements in the deep learning model’s performance compared to conventional methods. Notably, the ResNet50 model delivers the best performance with accuracy rates of 90.73% using NIGHT-VIS-CLAHE data, demonstrating a 16% improvement in accuracy. For benchmark purposes, the InceptionV3, GoogleNet, and MobileNetV2 models also show enhanced accuracy through CLAHE implementation. Furthermore, NIGHT-VIS-CLAHE implementation in ResNet50 achieved 90.29% accuracy, surpassing the best NIGHT-IR InceptionV3 at 89.27%, highlighting the advantage of ResNet50 with CLAHE in low-light conditions even against infra-red sensor.
- Open Access
- 15 May 2023
- Published by Akademia Baru Publishing in Journal of Advanced Research in Applied Sciences and Engineering Technology
- Vol. 30 (3) , 222-235
Human Detection from Drone using You Only Look Once (YOLOv5) for Search and Rescue Operation
Drones are unmanned aerial vehicles that can be remotely operated to perform a variety of tasks. They have been used in search and rescue operations since the early 2000s and have proven to be invaluable tools for quickly locating missing persons in difficult terrain and environment. In certain cases, automated human detection on drone camera feed can help the responder to locate the victims more effectively. In this work, we propose the use of a deep learning method called You Only Look Once version 5, or YOLOv5. The YOLOv5 model is trained using data collected during a simulation of search and rescue operations, where mannequins were used to represent human victims. Video was acquired using DJI Matrice 300 drone with Zenmuse H20T camera which flew around an area with various terrains such as farms, ravines, and river of more than 15,000 m2, at a height of 40 meters. The drone used grid, circular and zigzag flying patterns, with three different levels of camera zooms, and the data was captured on different days and times. The total duration of the video collected at 1080p@30fps is 148 minutes 26 seconds. Five pretrained models of YOLOv5 with different complexities were trained and tested using this dataset. Results showed that pretrained yolov5l6 model delivered the best precision, recall and mAP50 rate at 0.668, 0.303 and 0.346 respectively. Besides, the experiment also showed that we can improve the overall performance by using images acquired at 6x zoom magnification level where precision, recall, and mAP50 rate are increased to 0.846, 0.543, and 0.591 respectively. yolov5l6 model also delivered an acceptable inference time of 43ms per 1920x1080 resolution image, thus it can run at a respectable 23fps.
- 15 May 2023
- Published by Akademia Baru Publishing in Journal of Advanced Research in Applied Sciences and Engineering Technology
- Vol. 30 (3) , 222-235
Open AccessHuman Detection from Drone using You Only Look Once (YOLOv5) for Search and Rescue Operation
Drones are unmanned aerial vehicles that can be remotely operated to perform a variety of tasks. They have been used in search and rescue operations since the early 2000s and have proven to be invaluable tools for quickly locating missing persons in difficult terrain and environment. In certain cases, automated human detection on drone camera feed can help the responder to locate the victims more effectively. In this work, we propose the use of a deep learning method called You Only Look Once version 5, or YOLOv5. The YOLOv5 model is trained using data collected during a simulation of search and rescue operations, where mannequins were used to represent human victims. Video was acquired using DJI Matrice 300 drone with Zenmuse H20T camera which flew around an area with various terrains such as farms, ravines, and river of more than 15,000 m2, at a height of 40 meters. The drone used grid, circular and zigzag flying patterns, with three different levels of camera zooms, and the data was captured on different days and times. The total duration of the video collected at 1080p@30fps is 148 minutes 26 seconds. Five pretrained models of YOLOv5 with different complexities were trained and tested using this dataset. Results showed that pretrained yolov5l6 model delivered the best precision, recall and mAP50 rate at 0.668, 0.303 and 0.346 respectively. Besides, the experiment also showed that we can improve the overall performance by using images acquired at 6x zoom magnification level where precision, recall, and mAP50 rate are increased to 0.846, 0.543, and 0.591 respectively. yolov5l6 model also delivered an acceptable inference time of 43ms per 1920x1080 resolution image, thus it can run at a respectable 23fps.