Muhammed Swavaf
Current Organizations / Institutes
The organizations/institutes from the scholar's latest article
Research Subjects
Muhammed Swavaf has a total of 22 co-authors. Here are the top co-authors by number of shared publications:
Fady Alnajjar
United Arab Emirates University,Dept. Computer Science and Software Engineering,Al Ain,UAE
3 shared publications- Fady Alnajjar
College of Information Technology, United Arab Emirates University, Al Ain 15551, United Arab Emirates
3 shared publications Luqman Ali
Emirates Centre for Mobility Research United Arab Emirates University,Dept. Computer Science and Software Engineering,Al Ain,UAE
3 shared publicationsMedha Mohan Ambali Parambil
United Arab Emirates University,Dept. Computer Science and Software Engineering,Al Ain,UAE
3 shared publicationsHamad Aljassmi
Emirates Centre for Mobility Research,Dept. Civil and Enviromental Engieering Engieering,Al Ain,UAE
2 shared publications- Abdullah Al Mahmud
Swinburne University of Technology
1 shared publication
Latest Publications
- Open Access
- 4 September 2026
- Published by Elsevier in SSRN Electronic Journal
Comparative Evaluation of YOLO-Based Object Detection Models for Aerial Drone Imagery Analysis
A key component of real-time computer vision applications in fields like robotics, autonomous vehicles, and aerial surveillance is the You Only Look Once (YOLO) object detection framework. Despite ongoing research, there remains a gap in understanding the suitability of different YOLO model versions for domain-specific tasks, particularly in constrained computational-resource scenarios and high-object-variability settings, such as drone-based imagery. This study presents a comprehensive comparative analysis of YOLOv8 and YOLOv11 series, focusing on object detection in aerial drone imagery using the VisDrone dataset. We evaluated multiple variants of both YOLOv8 and YOLOv11 under consistent training settings. Our analysis considers key performance metrics, including mean average precision (mAP@50), inference speed (FPS), computational complexity (FLOPs), and model size. We also examine object-specific detection performance across ten annotated categories: pedestrian, people, bicycle, car, van, truck, tricycle, awning-tricycle, bus, and motor. The results indicate that YOLOv11 models offer modest improvements in accuracy, particularly for small and occluded objects such as pedestrians and bicycles, but at the cost of increased computational complexity (up to 45\% more FLOPs) and larger model sizes. YOLOv8 variants, in contrast, achieve better speed--accuracy trade-offs and are more suitable for real-time inference on resource-constrained platforms, with up to 25\% higher FPS in edge deployment scenarios. Our findings challenge the prevailing assumption that newer models are inherently superior and instead highlight the importance of selecting detection models based on task-specific requirements. This work provides practical guidance for researchers and practitioners in choosing the most appropriate YOLO model for object detection in aerial imagery.
- 4 September 2026
- Published by Elsevier in SSRN Electronic Journal
Open AccessComparative Evaluation of YOLO-Based Object Detection Models for Aerial Drone Imagery Analysis
A key component of real-time computer vision applications in fields like robotics, autonomous vehicles, and aerial surveillance is the You Only Look Once (YOLO) object detection framework. Despite ongoing research, there remains a gap in understanding the suitability of different YOLO model versions for domain-specific tasks, particularly in constrained computational-resource scenarios and high-object-variability settings, such as drone-based imagery. This study presents a comprehensive comparative analysis of YOLOv8 and YOLOv11 series, focusing on object detection in aerial drone imagery using the VisDrone dataset. We evaluated multiple variants of both YOLOv8 and YOLOv11 under consistent training settings. Our analysis considers key performance metrics, including mean average precision (mAP@50), inference speed (FPS), computational complexity (FLOPs), and model size. We also examine object-specific detection performance across ten annotated categories: pedestrian, people, bicycle, car, van, truck, tricycle, awning-tricycle, bus, and motor. The results indicate that YOLOv11 models offer modest improvements in accuracy, particularly for small and occluded objects such as pedestrians and bicycles, but at the cost of increased computational complexity (up to 45\% more FLOPs) and larger model sizes. YOLOv8 variants, in contrast, achieve better speed--accuracy trade-offs and are more suitable for real-time inference on resource-constrained platforms, with up to 25\% higher FPS in edge deployment scenarios. Our findings challenge the prevailing assumption that newer models are inherently superior and instead highlight the importance of selecting detection models based on task-specific requirements. This work provides practical guidance for researchers and practitioners in choosing the most appropriate YOLO model for object detection in aerial imagery.
- Open Access
- 12 March 2026
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Access
- Vol. 14, 41529-41540
Multi-Temporal Feature Integration: Learning Complementary Motion and Semantic Representations for Gait Recognition
Gait recognition is a promising biometric modality for the non-invasive, long-distance identification of individuals based on their unique walking patterns. Despite its potential advantages, real-world gait recognition is hindered by the unconstrained nature of gait acquisition, where factors such as camera viewpoint, clothing, and carried objects can distort body shape and gait dynamics. While most existing methods rely on 3D convolutions to extract spatiotemporal features, they typically operate at a fixed temporal resolution (TR), which limits their ability to jointly model fine-grained motion details and high-level spatial semantics. To overcome this limitation, we propose a novel method comprising two modules with a dual TR gait recognition strategy that integrates complementary features extracted from high TR (HTR) and low TR (LTR) video streams. The HTR feature extraction module captures subtle motion dynamics, while the LTR feature extraction module preserves rich spatial and semantic shape details. Additionally, an effective lateral fusion mechanism is employed to exploit multi-scale temporal information, enabling more discriminative feature representation for gait recognition. By fusing both features, the proposed method demonstrates strong robustness across diverse covariate conditions, including camera-view variations, carried objects, clothing variations, and real-world scenarios. Extensive experiments on four publicly available benchmark datasets (CASIA-B, CCPG, OULP, and Gait3D) validate the effectiveness of our proposed method, which achieves superior accuracy. On the CASIA-B dataset, the proposed method achieves average rank-1 accuracies of 98.9%, 96.2%, and 86.7% for the normal walking, carried object, and clothing variation covariate conditions, respectively. Our method also achieves average rank-1 accuracy on each dataset of 81.2% for CCPG, 99.1% for OULP, and 54.2% for Gait3D. The source code will be available at https://github.com/MDROKONMIA/HTR_LTR/.
- 12 March 2026
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Access
- Vol. 14, 41529-41540
Open AccessMulti-Temporal Feature Integration: Learning Complementary Motion and Semantic Representations for Gait Recognition
Gait recognition is a promising biometric modality for the non-invasive, long-distance identification of individuals based on their unique walking patterns. Despite its potential advantages, real-world gait recognition is hindered by the unconstrained nature of gait acquisition, where factors such as camera viewpoint, clothing, and carried objects can distort body shape and gait dynamics. While most existing methods rely on 3D convolutions to extract spatiotemporal features, they typically operate at a fixed temporal resolution (TR), which limits their ability to jointly model fine-grained motion details and high-level spatial semantics. To overcome this limitation, we propose a novel method comprising two modules with a dual TR gait recognition strategy that integrates complementary features extracted from high TR (HTR) and low TR (LTR) video streams. The HTR feature extraction module captures subtle motion dynamics, while the LTR feature extraction module preserves rich spatial and semantic shape details. Additionally, an effective lateral fusion mechanism is employed to exploit multi-scale temporal information, enabling more discriminative feature representation for gait recognition. By fusing both features, the proposed method demonstrates strong robustness across diverse covariate conditions, including camera-view variations, carried objects, clothing variations, and real-world scenarios. Extensive experiments on four publicly available benchmark datasets (CASIA-B, CCPG, OULP, and Gait3D) validate the effectiveness of our proposed method, which achieves superior accuracy. On the CASIA-B dataset, the proposed method achieves average rank-1 accuracies of 98.9%, 96.2%, and 86.7% for the normal walking, carried object, and clothing variation covariate conditions, respectively. Our method also achieves average rank-1 accuracy on each dataset of 81.2% for CCPG, 99.1% for OULP, and 54.2% for Gait3D. The source code will be available at https://github.com/MDROKONMIA/HTR_LTR/.
- 16 December 2024
- Published by Institute of Electrical and Electronics Engineers (IEEE)
- p. 178-183
SWIN and Vision Transformer-Driven Crack Detection in the Al Qattara Oasis, UAE: Towards Sustainable Infrastructure Management
Heritage sites are central to the cultural identity and historical narrative of a community. The Al Qattara Oasis, located in the United Arab Emirates (UAE), illustrates this function well. This study addresses the pressing need to preserve and maintain built heritage by employing modern technological solutions for infrastructure upkeep. Specifically, it focuses on crack detection, a critical aspect of ensuring the structural integrity of heritage buildings. Utilizing data collected from various structures across the UAE, obtained through handheld cameras, the effectiveness of Vision Transformers (ViTs) and Swin Transformers in identifying cracks within Al Qattara is assessed. The research thoroughly evaluates models, with particular attention to different input patch sizes. Through systematic experimentation and analysis over 100 epochs, ViT models with a patch size 16 exhibit significant promise in crack detection. Notably, the ViT-16 model achieves good performance metrics, including training and validation accuracies of 85% and (82%) respectively, and correctly classifies 791 out of 957 cracks and 772 out of 903 non-cracks. In contrast, Swin Transformers show lower validation accuracies (70-74%) and higher misclassification rates. The outcomes underscore the potential of ViT models in enhancing infrastructure maintenance efforts within heritage sites like Qattara Oasis. By combining advanced technology with a profound respect for historical preservation, this study aims to contribute to the sustainable conservation and protection of cultural heritage, ensuring that future generations can continue to appreciate the enduring legacy of sites such as Qattara Oasis.
- 16 December 2024
- Published by Institute of Electrical and Electronics Engineers (IEEE)
- p. 178-183
SWIN and Vision Transformer-Driven Crack Detection in the Al Qattara Oasis, UAE: Towards Sustainable Infrastructure Management
Heritage sites are central to the cultural identity and historical narrative of a community. The Al Qattara Oasis, located in the United Arab Emirates (UAE), illustrates this function well. This study addresses the pressing need to preserve and maintain built heritage by employing modern technological solutions for infrastructure upkeep. Specifically, it focuses on crack detection, a critical aspect of ensuring the structural integrity of heritage buildings. Utilizing data collected from various structures across the UAE, obtained through handheld cameras, the effectiveness of Vision Transformers (ViTs) and Swin Transformers in identifying cracks within Al Qattara is assessed. The research thoroughly evaluates models, with particular attention to different input patch sizes. Through systematic experimentation and analysis over 100 epochs, ViT models with a patch size 16 exhibit significant promise in crack detection. Notably, the ViT-16 model achieves good performance metrics, including training and validation accuracies of 85% and (82%) respectively, and correctly classifies 791 out of 957 cracks and 772 out of 903 non-cracks. In contrast, Swin Transformers show lower validation accuracies (70-74%) and higher misclassification rates. The outcomes underscore the potential of ViT models in enhancing infrastructure maintenance efforts within heritage sites like Qattara Oasis. By combining advanced technology with a profound respect for historical preservation, this study aims to contribute to the sustainable conservation and protection of cultural heritage, ensuring that future generations can continue to appreciate the enduring legacy of sites such as Qattara Oasis.
- Open Access
- 6 August 2024
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Access
- Vol. 12, 109427-109442
Navigating the YOLO Landscape: A Comparative Study of Object Detection Models for Emotion Recognition
The You Only Look Once (YOLO) series, renowned for its efficiency and versatility in object detection, has become a fundamental component in diverse fields ranging from autonomous vehicles to robotics and video surveillance. Despite its widespread application, a notable gap exists in the literature concerning selecting YOLO models for specific tasks. Current trends often lean towards the latest models, potentially overlooking crucial factors such as computational complexity, speed, accuracy, model size, adaptability, and generalization. This approach may not always yield the optimal choice for a given application. Therefore, this paper aims to provide an exhaustive comparative analysis of various YOLO models, focusing on emotion recognition. We trained and tested YOLOv5, YOLOv7, YOLOv8, and YOLOv9 along with their respective variants, using a subset of AffectNet dataset, which consists of facial images annotated with one of five emotions, namely angry, happy, sad, neutral, and surprise. The study evaluates the models based on several key parameters: accuracy using metrics like mean Average Precision (mAP), inference time and FPS, model size, adaptability to altered datasets, and generalization capability. Comprehensive results are presented, highlighting the strengths and limitations of each model variant across these parameters. Insights are provided to guide researchers in selecting the most suitable YOLO architecture for their specific emotion recognition requirements, considering factors such as computational constraints, real-time performance needs, and the importance of accuracy vs efficiency tradeoffs. The analysis reveals the exceptional performances of certain models like YOLOv9e for high accuracy and YOLOv8n for balancing speed and accuracy. Overall, this work fills a crucial gap by offering a detailed comparative study to facilitate informed decision-making when deploying YOLO for facial emotion recognition tasks.
- 6 August 2024
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Access
- Vol. 12, 109427-109442
Open AccessNavigating the YOLO Landscape: A Comparative Study of Object Detection Models for Emotion Recognition
The You Only Look Once (YOLO) series, renowned for its efficiency and versatility in object detection, has become a fundamental component in diverse fields ranging from autonomous vehicles to robotics and video surveillance. Despite its widespread application, a notable gap exists in the literature concerning selecting YOLO models for specific tasks. Current trends often lean towards the latest models, potentially overlooking crucial factors such as computational complexity, speed, accuracy, model size, adaptability, and generalization. This approach may not always yield the optimal choice for a given application. Therefore, this paper aims to provide an exhaustive comparative analysis of various YOLO models, focusing on emotion recognition. We trained and tested YOLOv5, YOLOv7, YOLOv8, and YOLOv9 along with their respective variants, using a subset of AffectNet dataset, which consists of facial images annotated with one of five emotions, namely angry, happy, sad, neutral, and surprise. The study evaluates the models based on several key parameters: accuracy using metrics like mean Average Precision (mAP), inference time and FPS, model size, adaptability to altered datasets, and generalization capability. Comprehensive results are presented, highlighting the strengths and limitations of each model variant across these parameters. Insights are provided to guide researchers in selecting the most suitable YOLO architecture for their specific emotion recognition requirements, considering factors such as computational constraints, real-time performance needs, and the importance of accuracy vs efficiency tradeoffs. The analysis reveals the exceptional performances of certain models like YOLOv9e for high accuracy and YOLOv8n for balancing speed and accuracy. Overall, this work fills a crucial gap by offering a detailed comparative study to facilitate informed decision-making when deploying YOLO for facial emotion recognition tasks.
- 3 December 2023
- Published by Springer Nature
- p. 13-21
Human Perception of Emotional Responses to Changes in Auditory Attributes of Humanoid Agents
- 3 December 2023
- Published by Springer Nature
- p. 13-21
Human Perception of Emotional Responses to Changes in Auditory Attributes of Humanoid Agents
- 14 November 2023
- Published by Institute of Electrical and Electronics Engineers (IEEE)
- p. 262-267
Visualizing Student Attention in Smart Classrooms: A Preliminary Study
In recent years, there has been a swift progression in employing novel methods in classrooms to enhance students’ academic achievements, especially in line with the growing digitization of education. Such methods often encompass systems like facial recognition to monitor various aspects of students, including attendance, emotional states, and attention. These tools are capable of evaluating students’ presence and engagement in class, offering quantifiable metrics regarding their concentration and emotions. However, a prominent challenge has been the translation of this data into an accessible form that enables educators to assess and enhance their teaching techniques swiftly. Our suggested solution tackles this issue by offering a real-time visual depiction of students’ classroom status through different visualization techniques. These visual aids allow teachers to promptly recognize trends in student focus, thus aiding in the strategic alteration of teaching styles. Furthermore, these visual representations can be tailored to display various metrics and applied to tasks beyond monitoring attention, like overseeing attendance or assessing student progress. By integrating these advanced visualizations into the educational process, both teaching efficacy and the learning experience for students and teachers alike can be substantially elevated.
- 14 November 2023
- Published by Institute of Electrical and Electronics Engineers (IEEE)
- p. 262-267
Visualizing Student Attention in Smart Classrooms: A Preliminary Study
In recent years, there has been a swift progression in employing novel methods in classrooms to enhance students’ academic achievements, especially in line with the growing digitization of education. Such methods often encompass systems like facial recognition to monitor various aspects of students, including attendance, emotional states, and attention. These tools are capable of evaluating students’ presence and engagement in class, offering quantifiable metrics regarding their concentration and emotions. However, a prominent challenge has been the translation of this data into an accessible form that enables educators to assess and enhance their teaching techniques swiftly. Our suggested solution tackles this issue by offering a real-time visual depiction of students’ classroom status through different visualization techniques. These visual aids allow teachers to promptly recognize trends in student focus, thus aiding in the strategic alteration of teaching styles. Furthermore, these visual representations can be tailored to display various metrics and applied to tasks beyond monitoring attention, like overseeing attendance or assessing student progress. By integrating these advanced visualizations into the educational process, both teaching efficacy and the learning experience for students and teachers alike can be substantially elevated.
- 7 July 2023
- Published by International Association for Automation and Robotics in Construction (IAARC) in Proceedings of the 37th International Symposium on Automation and Robotics in Construction (isarc)
- p. 699-705
Crack Detection and Localization in Stone Floor Tiles using Vision Transformer approach
- 7 July 2023
- Published by International Association for Automation and Robotics in Construction (IAARC) in Proceedings of the 37th International Symposium on Automation and Robotics in Construction (isarc)
- p. 699-705
Crack Detection and Localization in Stone Floor Tiles using Vision Transformer approach