Eko Mulyanto Yuniarno
Current Organizations / Institutes
The organizations/institutes from the scholar's latest article
Research Subjects
Eko Mulyanto Yuniarno has a total of 119 co-authors. Here are the top co-authors by number of shared publications:
- Mauridhi Hery Purnomo
Department of Electrical Engineering, Institut Teknologi Sepuluh Nopember, Surabaya, Indonesia
21 shared publications - I. Ketut Eddy Purnama
Department of Electrical Engineering, Institut Teknologi Sepuluh Nopember, Surabaya, Indonesia
13 shared publications - Supeno Mardi Susiki Nugroho
Department of Computer Engineering, Institut Teknologi Sepuluh Nopember, Surabaya, Indonesia
7 shared publications - Wiwik Anggraeni
Institut Teknologi Sepuluh Nopember
5 shared publications Mochamad Hariadi
Institut Technology of Sepuluh November Surabaya
4 shared publications- Reza Fuad Rachmadi
Department of Computer Engineering, Institut Teknologi Sepuluh Nopember, Surabaya, Indonesia
4 shared publications
Latest Publications
- Open Access
- 2 October 2026
- Published by Engineering, Technology & Applied Science Research in Engineering, Technology & Applied Science Research
- Vol. 16  (5) , 39788-39796
Multi-Scale CNN-LSTM with Atrous Spatial Pyramid Pooling for Enhanced Brain Tumor Classification in ADC-MRI Sequences
Discriminating between malignant and benign brain tumors using ADC-MRI sequences remains challenging due to high similarity in morphology, intra-class heterogeneity, and the sequential nature of the neuroimaging data. This paper introduces a Multi-Scale CNN-LSTM network enhanced by an Atrous Spatial Pyramid Pooling (ASPP) mechanism, bi-directional temporal processing, and squeeze-and-excitation channel attention, which, to the best of our knowledge, is the first model combining multi-scale contextual information with temporal dependency modeling for brain tumor classification using ADC-MRIs. Specifically, local-to-global feature extraction is performed using parallel Inception-based convolutional branches, ASPP is used with variable dilation rates to expand the receptive field without reducing spatial resolution, while Bi-LSTM learns both forward and backward sequential dependencies over seven-slice subsequences of ADC MRIs. A focal loss function was used during training to overcome the issue of imbalanced classes. The proposed model was evaluated using an ADC-MRI dataset of 1,202 images from 41 patients. Results show 98.05% accuracy, 100.00% sensitivity, 95.52% specificity, 96.67% precision, F1-score of 98.31%, AUC-ROC of 0.9993, and MCC of 0.9609. Seven variants ablation experiments prove that ASPP integration provides the largest single-component improvement of 5.89 percentage points (pp) in accuracy compared to the baseline CNN-LSTM, while focal loss gives zero false negative predictions.
- 2 October 2026
- Published by Engineering, Technology & Applied Science Research in Engineering, Technology & Applied Science Research
- Vol. 16  (5) , 39788-39796
Open AccessMulti-Scale CNN-LSTM with Atrous Spatial Pyramid Pooling for Enhanced Brain Tumor Classification in ADC-MRI Sequences
Discriminating between malignant and benign brain tumors using ADC-MRI sequences remains challenging due to high similarity in morphology, intra-class heterogeneity, and the sequential nature of the neuroimaging data. This paper introduces a Multi-Scale CNN-LSTM network enhanced by an Atrous Spatial Pyramid Pooling (ASPP) mechanism, bi-directional temporal processing, and squeeze-and-excitation channel attention, which, to the best of our knowledge, is the first model combining multi-scale contextual information with temporal dependency modeling for brain tumor classification using ADC-MRIs. Specifically, local-to-global feature extraction is performed using parallel Inception-based convolutional branches, ASPP is used with variable dilation rates to expand the receptive field without reducing spatial resolution, while Bi-LSTM learns both forward and backward sequential dependencies over seven-slice subsequences of ADC MRIs. A focal loss function was used during training to overcome the issue of imbalanced classes. The proposed model was evaluated using an ADC-MRI dataset of 1,202 images from 41 patients. Results show 98.05% accuracy, 100.00% sensitivity, 95.52% specificity, 96.67% precision, F1-score of 98.31%, AUC-ROC of 0.9993, and MCC of 0.9609. Seven variants ablation experiments prove that ASPP integration provides the largest single-component improvement of 5.89 percentage points (pp) in accuracy compared to the baseline CNN-LSTM, while focal loss gives zero false negative predictions.
- Open Access
- 15 May 2026
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Access
- Vol. 14, 76722-76747
MyoScarNet-RA: A Residual Attention Network for Myocardial Surface Area Quantification Based on Scar Segmentation
Myocardial infarction (MI) often leads to fibrotic scar tissue formation, significantly increasing the risk of heart failure and recurrent infarctions. Accurate segmentation and quantification of myocardial scars are essential for early diagnosis, risk stratification, and effective treatment planning. This study presents MyoScarNet-RA, an advanced deep learning model designed for high-quality myocardial scar segmentation and automated quantification from multi-sequence cardiac magnetic resonance imaging (MRI). MyoScarNet-RA integrates a residual attention mechanism within a hierarchical encoder-decoder architecture, enabling precise delineation of myocardial scar regions through synergistic combination of residual learning and hybrid channel-spatial attention. Trained and evaluated on the MyoPS2020 dataset with patient-level data splitting, MyoScarNet-RA demonstrates superior boundary localization performance, achieving a Dice score of 0.9326±0.0368, IoU of 0.8758±0.0625, HD of 1.3117±0.5287, and SSIM of 0.9406±0.0301. The exceptional HD performance highlights its strength in accurate scar boundary detection. Beyond segmentation, the validated quantification pipeline achieves mean absolute error of 5.43 mm2 and strong correlation (r=0.952, p2, 95% limits: −6.77 to +14.55 mm2). With efficient GPU memory utilization (0.40 GB), MyoScarNet-RA offers a practical automated solution for cardiovascular risk assessment and improved clinical decision-making.
- 15 May 2026
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Access
- Vol. 14, 76722-76747
Open AccessMyoScarNet-RA: A Residual Attention Network for Myocardial Surface Area Quantification Based on Scar Segmentation
Myocardial infarction (MI) often leads to fibrotic scar tissue formation, significantly increasing the risk of heart failure and recurrent infarctions. Accurate segmentation and quantification of myocardial scars are essential for early diagnosis, risk stratification, and effective treatment planning. This study presents MyoScarNet-RA, an advanced deep learning model designed for high-quality myocardial scar segmentation and automated quantification from multi-sequence cardiac magnetic resonance imaging (MRI). MyoScarNet-RA integrates a residual attention mechanism within a hierarchical encoder-decoder architecture, enabling precise delineation of myocardial scar regions through synergistic combination of residual learning and hybrid channel-spatial attention. Trained and evaluated on the MyoPS2020 dataset with patient-level data splitting, MyoScarNet-RA demonstrates superior boundary localization performance, achieving a Dice score of 0.9326±0.0368, IoU of 0.8758±0.0625, HD of 1.3117±0.5287, and SSIM of 0.9406±0.0301. The exceptional HD performance highlights its strength in accurate scar boundary detection. Beyond segmentation, the validated quantification pipeline achieves mean absolute error of 5.43 mm2 and strong correlation (r=0.952, p2, 95% limits: −6.77 to +14.55 mm2). With efficient GPU memory utilization (0.40 GB), MyoScarNet-RA offers a practical automated solution for cardiovascular risk assessment and improved clinical decision-making.
- Open Access
- 1 January 2026
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Access
- Vol. 14, 142999-143021
Fine-Tuning UniXcoder for Code Smell Detection in Java Projects
- 1 January 2026
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Access
- Vol. 14, 142999-143021
Open AccessFine-Tuning UniXcoder for Code Smell Detection in Java Projects
- Open Access
- 8 December 2025
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Access
- Vol. 13, 208590-208611
PEAK-U-Net: A Progressive Edge-Aware Attention KAN U-Net With Sliding Window for Crack Segmentation
Crack detection and quantification in concrete structures remains a critical challenge in structural health monitoring. Traditional manual inspection methods are time-consuming and prone to human error, while existing deep learning approaches struggle with fine-grained edge detection and accurate dimensional quantification. This paper introduces PEAK-U-Net (Progressive Edge-Aware Attention Kolmogorov-Arnold U-Net), a novel architecture addressing these limitations through three key innovations. First, we propose a progressive edge-aware attention mechanism that adaptively focuses on crack boundaries across multiple scales. Second, we present the integration of Kolmogorov-Arnold Networks (KAN) into U-Net architecture, replacing traditional MLP layers with learnable activation functions for superior feature approximation. Third, we develop a multi-scale sliding window framework maintaining sub-millimeter measurement precision based on calibrated pixel resolution. The PEAK-U-Net employs an encoder-decoder structure with enhanced KAN modules in bottleneck and decoder sections, incorporating attention gates and edge-aware convolutions. Our sliding window approach processes images through systematic patch extraction and reconstructs results with maintained spatial accuracy for precise crack width measurements. Extensive experimental validation demonstrates that PEAK-U-Net achieves state-of-the-art performance, particularly excelling in small crack detection with 99.45% accuracy, 81.85% DICE coefficient, and 74.75% Mean IOU on the CrackForest dataset. The method demonstrates superior precision (84.31%) and recall (82.36%) with the lowest loss value (0.21), while successfully classifying crack severity from Hairline (0.07 mm) to Medium (0.20-0.25 mm). PEAK-U-Net represents a transformative contribution to computer vision-based structural health monitoring, offering significant potential to enhance preventive maintenance strategies and provide a robust foundation for next-generation automated inspection systems.
- 8 December 2025
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Access
- Vol. 13, 208590-208611
Open AccessPEAK-U-Net: A Progressive Edge-Aware Attention KAN U-Net With Sliding Window for Crack Segmentation
Crack detection and quantification in concrete structures remains a critical challenge in structural health monitoring. Traditional manual inspection methods are time-consuming and prone to human error, while existing deep learning approaches struggle with fine-grained edge detection and accurate dimensional quantification. This paper introduces PEAK-U-Net (Progressive Edge-Aware Attention Kolmogorov-Arnold U-Net), a novel architecture addressing these limitations through three key innovations. First, we propose a progressive edge-aware attention mechanism that adaptively focuses on crack boundaries across multiple scales. Second, we present the integration of Kolmogorov-Arnold Networks (KAN) into U-Net architecture, replacing traditional MLP layers with learnable activation functions for superior feature approximation. Third, we develop a multi-scale sliding window framework maintaining sub-millimeter measurement precision based on calibrated pixel resolution. The PEAK-U-Net employs an encoder-decoder structure with enhanced KAN modules in bottleneck and decoder sections, incorporating attention gates and edge-aware convolutions. Our sliding window approach processes images through systematic patch extraction and reconstructs results with maintained spatial accuracy for precise crack width measurements. Extensive experimental validation demonstrates that PEAK-U-Net achieves state-of-the-art performance, particularly excelling in small crack detection with 99.45% accuracy, 81.85% DICE coefficient, and 74.75% Mean IOU on the CrackForest dataset. The method demonstrates superior precision (84.31%) and recall (82.36%) with the lowest loss value (0.21), while successfully classifying crack severity from Hairline (0.07 mm) to Medium (0.20-0.25 mm). PEAK-U-Net represents a transformative contribution to computer vision-based structural health monitoring, offering significant potential to enhance preventive maintenance strategies and provide a robust foundation for next-generation automated inspection systems.
- Open Access
- 30 October 2025
- Published by Serious Games Society in International Journal of Serious Games
- Vol. 12  (4) , 217-238
Player Performance Estimation Through Adaptive Puzzle Levels Generation in Game-Based Assessment
Estimating players’ abilities without compromising alignment with assessment objectives remains a key challenge in developing Game-Based Assessment (GBA). This study proposes a method to estimate player performance through the procedural generation of puzzle game levels that adapt to player achievement. The method integrates Dynamic Difficulty Adjustment (DDA) and player modeling to dynamically align content difficulty with individual abilities while maintaining assessment validity. Unlike most previous PCG-GBA studies, which determine difficulty solely from technical parameters, the proposed method combines cognitive (arithmetic content) and technical (puzzle parameters) indicators, validated through paper-based tests. Implemented in a basic arithmetic puzzle game with sixth-grade primary school students in Indonesia, the approach generated 1,379 gameplay level records. Statistical analysis shows a strong positive correlation between paper-based test scores and the average difficulty of completed levels (r = 0.734, p < 0.001). A low absolute difference between difficulty level and player performance (M = 0.012, SD = 0.028) and a small RMSE (0.031) confirm the accuracy and consistency of difficulty adaptation. The proposed method enables accurate performance estimation through adaptive gameplay, contributing to the development and external validation of PCG-based GBA and pointing toward its potential scalability for personalized educational assessments.
- 30 October 2025
- Published by Serious Games Society in International Journal of Serious Games
- Vol. 12  (4) , 217-238
Open AccessPlayer Performance Estimation Through Adaptive Puzzle Levels Generation in Game-Based Assessment
Estimating players’ abilities without compromising alignment with assessment objectives remains a key challenge in developing Game-Based Assessment (GBA). This study proposes a method to estimate player performance through the procedural generation of puzzle game levels that adapt to player achievement. The method integrates Dynamic Difficulty Adjustment (DDA) and player modeling to dynamically align content difficulty with individual abilities while maintaining assessment validity. Unlike most previous PCG-GBA studies, which determine difficulty solely from technical parameters, the proposed method combines cognitive (arithmetic content) and technical (puzzle parameters) indicators, validated through paper-based tests. Implemented in a basic arithmetic puzzle game with sixth-grade primary school students in Indonesia, the approach generated 1,379 gameplay level records. Statistical analysis shows a strong positive correlation between paper-based test scores and the average difficulty of completed levels (r = 0.734, p < 0.001). A low absolute difference between difficulty level and player performance (M = 0.012, SD = 0.028) and a small RMSE (0.031) confirm the accuracy and consistency of difficulty adaptation. The proposed method enables accurate performance estimation through adaptive gameplay, contributing to the development and external validation of PCG-based GBA and pointing toward its potential scalability for personalized educational assessments.
- Open Access
- 1 August 2025
- Published by Lembaga Penelitian dan Pengabdian kepada Masyarakat ITS in JAREE (Journal on Advanced Research in Electrical Engineering)
- Vol. 9  (2)
Markerless Facial Reconstruction Motion Capture Using Triangulation Method
Motion capture is a popular research topic, with one of its main applications being human face reconstruction. The demand for converting 2D images into 3D reconstructions continues to increase, especially in facial reconstruction, where progress is made in improving the accuracy of facial position prediction. However, there is still a significant gap in developing facial reconstruction technologies that can consistently convert 2D to 3D data with high accuracy, especially in scenarios involving dynamic facial expressions, diverse facial angles, and complex environmental conditions. Therefore, an approach using the triangulation method for 3D face reconstruction in the real world was developed. In the experiments, two cameras were used to obtain two face landmark coordinates so that the triangulation method can be implemented for 3D face reconstruction. This research aims to develop a motion capture approach that is able to accurately and efficiently transform 2D data into 3D face models without the need for complex hardware. The main contribution of this research is the development of a machine learning-based markerless motion capture technique designed to improve the accuracy of face position prediction in 3D face reconstruction from 2D data in realistic environments. This method seeks to bridge the current technology gap by providing a more flexible and reliable solution, expanding the potential applications of motion capture in various fields without dependence on specialized hardware. The results of face reconstruction research using markerless motion capture and triangulation method show RMSE values of 3.560839 for eyes, 1.644749 for nose, and 4.054638 for lips.
- 1 August 2025
- Published by Lembaga Penelitian dan Pengabdian kepada Masyarakat ITS in JAREE (Journal on Advanced Research in Electrical Engineering)
- Vol. 9  (2)
Open AccessMarkerless Facial Reconstruction Motion Capture Using Triangulation Method
Motion capture is a popular research topic, with one of its main applications being human face reconstruction. The demand for converting 2D images into 3D reconstructions continues to increase, especially in facial reconstruction, where progress is made in improving the accuracy of facial position prediction. However, there is still a significant gap in developing facial reconstruction technologies that can consistently convert 2D to 3D data with high accuracy, especially in scenarios involving dynamic facial expressions, diverse facial angles, and complex environmental conditions. Therefore, an approach using the triangulation method for 3D face reconstruction in the real world was developed. In the experiments, two cameras were used to obtain two face landmark coordinates so that the triangulation method can be implemented for 3D face reconstruction. This research aims to develop a motion capture approach that is able to accurately and efficiently transform 2D data into 3D face models without the need for complex hardware. The main contribution of this research is the development of a machine learning-based markerless motion capture technique designed to improve the accuracy of face position prediction in 3D face reconstruction from 2D data in realistic environments. This method seeks to bridge the current technology gap by providing a more flexible and reliable solution, expanding the potential applications of motion capture in various fields without dependence on specialized hardware. The results of face reconstruction research using markerless motion capture and triangulation method show RMSE values of 3.560839 for eyes, 1.644749 for nose, and 4.054638 for lips.
- Open Access
- 28 July 2025
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Access
- Vol. 13, 133465-133482
Revealing Depression Through Social Media via Adaptive Gated Cross-Modal Fusion Augmented With Insights From Personality Traits
The escalating prevalence of mental health disorders, particularly depression, underscores the urgent need for early and accurate detection systems. The widespread use of social media has produced an extensive repository of user-generated content, offering critical insight into individuals’ emotional and psychological states. This development enables novel approaches for automated depression detection. However, existing multimodal depression detection approaches often adopt rigid fusion strategies and disregard individual differences in expressive behavior by adopting generalized, one-size-fits-all frameworks. To bridge this gap, we introduce DeXMAG, a novel personalized depression detection framework that integrates a Cross-Modal Attention mechanism with an Adaptive Gated Fusion strategy. DeXMAG effectively combines textual content, visual imagery, and user-level personality traits to capture the heterogeneous and individualized patterns of depressive expression across modalities. Leveraging domain-specific transformer-based language models and a convolutional neural networks-based model, we extract modality-specific features and fuse them via a novel adaptive gating mechanism. Unlike traditional fusion techniques, our model leverages modality-aware attention signals and gating coefficients to enable flexible and personalized integration of features. An extensive ablation study reveals that the most substantial performance gain occurs when DeXMAG is augmented with insights from Myers–Briggs Type Indicator (MBTI) personality traits in conjunction with textual and visual features. Comprehensive experiments on multimodal social media datasets demonstrate that DeXMAG consistently outperforms unimodal baselines and state-of-the-art approaches, achieving an accuracy of 95.20% and an F1-score of 95.94%. These results reinforce the effectiveness of adaptive gated fusion and personality-aware modeling in advancing robust and personalized depression detection.
- 28 July 2025
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Access
- Vol. 13, 133465-133482
Open AccessRevealing Depression Through Social Media via Adaptive Gated Cross-Modal Fusion Augmented With Insights From Personality Traits
The escalating prevalence of mental health disorders, particularly depression, underscores the urgent need for early and accurate detection systems. The widespread use of social media has produced an extensive repository of user-generated content, offering critical insight into individuals’ emotional and psychological states. This development enables novel approaches for automated depression detection. However, existing multimodal depression detection approaches often adopt rigid fusion strategies and disregard individual differences in expressive behavior by adopting generalized, one-size-fits-all frameworks. To bridge this gap, we introduce DeXMAG, a novel personalized depression detection framework that integrates a Cross-Modal Attention mechanism with an Adaptive Gated Fusion strategy. DeXMAG effectively combines textual content, visual imagery, and user-level personality traits to capture the heterogeneous and individualized patterns of depressive expression across modalities. Leveraging domain-specific transformer-based language models and a convolutional neural networks-based model, we extract modality-specific features and fuse them via a novel adaptive gating mechanism. Unlike traditional fusion techniques, our model leverages modality-aware attention signals and gating coefficients to enable flexible and personalized integration of features. An extensive ablation study reveals that the most substantial performance gain occurs when DeXMAG is augmented with insights from Myers–Briggs Type Indicator (MBTI) personality traits in conjunction with textual and visual features. Comprehensive experiments on multimodal social media datasets demonstrate that DeXMAG consistently outperforms unimodal baselines and state-of-the-art approaches, achieving an accuracy of 95.20% and an F1-score of 95.94%. These results reinforce the effectiveness of adaptive gated fusion and personality-aware modeling in advancing robust and personalized depression detection.
- 12 March 2025
- Published by World Scientific Pub Co Pte Ltd in International Journal of Information Technology & Decision Making
- Vol. 24  (06) , 1685-1715
Predictive Analytics on Time Series Data To Generate A Deterministic Decision Model: A Case Study on School Reopening Safely During The Pandemic
This research focuses on developing a new decision-making model to evaluate school reopening strategies during the COVID-19 pandemic. The model integrates deep learning and factor analysis to address the urgent need to restart educational services without worsening the health crisis. It starts by gathering time series data from various districts to apply deep learning for predicting virus dynamics, emphasizing feature extraction and hyperparameter optimization. The subsequent phase involves factor analysis to discover key factors influencing virus spread, using outputs from the deep learning step. Based on these factors, clustering methods then sort districts into controllable or vulnerable groups. The final stage combines these analyzes into a deterministic decision model aiding policymakers in crafting school reopening guidelines. The model identifies three primary controllable factors: infection growth rate, reduction in active cases, and lowered mortality rates. Clustering then reveals that three groups are controllable, enabling specific interventions. This model is noteworthy for considering causal links between pandemic metrics and its adaptability to diverse datasets across districts/subdistricts, offering a scalable solution for decision-makers. The results highlight the importance of local infection trends and tailored data in shaping policies, showing that strong predictive analytics and insight into significant factors are crucial for developing effective, safe school reopening plans.
- 12 March 2025
- Published by World Scientific Pub Co Pte Ltd in International Journal of Information Technology & Decision Making
- Vol. 24  (06) , 1685-1715
Predictive Analytics on Time Series Data To Generate A Deterministic Decision Model: A Case Study on School Reopening Safely During The Pandemic
This research focuses on developing a new decision-making model to evaluate school reopening strategies during the COVID-19 pandemic. The model integrates deep learning and factor analysis to address the urgent need to restart educational services without worsening the health crisis. It starts by gathering time series data from various districts to apply deep learning for predicting virus dynamics, emphasizing feature extraction and hyperparameter optimization. The subsequent phase involves factor analysis to discover key factors influencing virus spread, using outputs from the deep learning step. Based on these factors, clustering methods then sort districts into controllable or vulnerable groups. The final stage combines these analyzes into a deterministic decision model aiding policymakers in crafting school reopening guidelines. The model identifies three primary controllable factors: infection growth rate, reduction in active cases, and lowered mortality rates. Clustering then reveals that three groups are controllable, enabling specific interventions. This model is noteworthy for considering causal links between pandemic metrics and its adaptability to diverse datasets across districts/subdistricts, offering a scalable solution for decision-makers. The results highlight the importance of local infection trends and tailored data in shaping policies, showing that strong predictive analytics and insight into significant factors are crucial for developing effective, safe school reopening plans.
- Open Access
- 3 February 2025
- Published by Lembaga Penelitian dan Pengabdian kepada Masyarakat ITS in JAREE (Journal on Advanced Research in Electrical Engineering)
- Vol. 9  (1)
Tracking Socer Player Based on Deepsort Algorithm with YOLOV8 FrameWork
—Tracking is a set procedure that entails assigninganidentificationtoacertainobjectandsubsequentlycon-sistently recognizing that object without altering the assignedidentification over a sequence of frame images and associatingitaccordingly.Whenperformingresearchonobjecttracking,especially in sports where the object of interest is a human, aresilient technology is necessary to facilitate the tracking process.When the state-of-the-art object detection approach, YOLOV8,is combined with the DeepSORT algorithm, it is anticipated toproduce highly accurate and exact outcomes in the trackingand detection of objects. Challenges in multi-object trackinginclude robustness, oculusion, and identity shifts. In our research,we take advantage of a fusion of YOLOV8 and DeepSORTalgorithmstoachieveahighlyreliableandprecisetrackingsolution. The implementation of the Kalman filter-based motionprediction in DeepSORT allows for the achievement of smoothtrajectories, whereas the YOLOV8 deep neural network usedassists in precisely recognizing the appearance of objects on thefield. The result of our experiment shown the tracking we get is38% HOTA, 47% DetA, 31% AssA, 68% DetPre, 35% AssRE,61% AssPr amd 79% LOcA.Index Terms—Tracking, DeepSORT, YOLO, MOT, Socce
- 3 February 2025
- Published by Lembaga Penelitian dan Pengabdian kepada Masyarakat ITS in JAREE (Journal on Advanced Research in Electrical Engineering)
- Vol. 9  (1)
Open AccessTracking Socer Player Based on Deepsort Algorithm with YOLOV8 FrameWork
—Tracking is a set procedure that entails assigninganidentificationtoacertainobjectandsubsequentlycon-sistently recognizing that object without altering the assignedidentification over a sequence of frame images and associatingitaccordingly.Whenperformingresearchonobjecttracking,especially in sports where the object of interest is a human, aresilient technology is necessary to facilitate the tracking process.When the state-of-the-art object detection approach, YOLOV8,is combined with the DeepSORT algorithm, it is anticipated toproduce highly accurate and exact outcomes in the trackingand detection of objects. Challenges in multi-object trackinginclude robustness, oculusion, and identity shifts. In our research,we take advantage of a fusion of YOLOV8 and DeepSORTalgorithmstoachieveahighlyreliableandprecisetrackingsolution. The implementation of the Kalman filter-based motionprediction in DeepSORT allows for the achievement of smoothtrajectories, whereas the YOLOV8 deep neural network usedassists in precisely recognizing the appearance of objects on thefield. The result of our experiment shown the tracking we get is38% HOTA, 47% DetA, 31% AssA, 68% DetPre, 35% AssRE,61% AssPr amd 79% LOcA.Index Terms—Tracking, DeepSORT, YOLO, MOT, Socce