Computer Vision-based Instance Segmentation and Data Augmentation for Construction Site Safety Compliance DOI

Kim Xyrus A. Dimaano,

Alvin Sarraga Alon

Опубликована: Дек. 6, 2023

Construction safety compliance requires modern technological innovations for effective monitoring and risk mitigation. This study presents an innovative approach addressing problems in construction environments that combines Transfer Learning-based Instance Segmentation Data Augmentation approaches with YOLOv8. The model succeeds at properly recognizing segmenting safety-critical objects within complex site images, a remarkable mean Average Precision (mAP) of 94.4%. key framework is YOLOv8, which well-known its real-time object recognition capabilities, allowing exact identification safety-related elements across diverse landscapes. Through transfer learning, the enhances capacity to identify distinguish by tailoring pre-existing knowledge intricacies scenarios. made more resilient adaptive advanced data augmentation techniques, guarantees it works well under variety environmental circumstances are common sites. By advancing computer vision technologies specifically designed applications high-risk work environments, this goals significantly advance enforcement laying groundwork better protocols preventive measures dynamic environment.

Язык: Английский

Integrating Drone Imagery and AI for Improved Construction Site Management through Building Information Modeling DOI Creative Commons
Wonjun Choi, Seunguk Na, Seokjae Heo

и другие.

Buildings, Год журнала: 2024, Номер 14(4), С. 1106 - 1106

Опубликована: Апрель 15, 2024

In the rapidly advancing field of construction, digital site management and Building Information Modeling (BIM) are pivotal. This study explores integration drone imagery into construction process, aiming to create BIM models with enhanced object recognition capabilities. Initially, research sought achieve photorealistic rendering point cloud (PCMs) using blur/sharpen filters generative adversarial network (GAN) models. However, these techniques did not fully meet desired outcomes for rendering. The then shifted investigating additional methods, such as fine-tuning algorithms real-world datasets, improve accuracy. study’s findings present a nuanced understanding limitations potential pathways achieving in PCM, underscoring complexity task laying groundwork future innovations this area. Although faced challenges attaining original goal detection, it contributes valuable insights that may inform technological development management.

Язык: Английский

Процитировано

11
A computer vision system for apple fruit sizing by means of low-cost depth camera and neural network application DOI Creative Commons
G. Bortolotti, Mirko Piani, M. L. Gullino

и другие.

Precision Agriculture, Год журнала: 2024, Номер 25(6), С. 2740 - 2757

Опубликована: Апрель 16, 2024

Abstract Fruit size is crucial for growers as it influences consumer willingness to buy and the price of fruit. growth along seasons are two parameters that can lead more precise orchard management favoring production sustainability. In this study, a Python-based computer vision system (CVS) sizing apples directly on tree was developed ease fruit tasks. The made consumer-grade depth camera tested at distances among 17 timings throughout season, in Fuji apple orchard. CVS exploited specifically trained YOLOv5 detection algorithm, circle trigonometric approach based information fruits. Comparisons with standard-trained models spherical objects were carried out. algorithm showed good performance, rate 92%. Good correlations ( r > 0.8) between estimated actual found. performance an overall mean error (mE) RMSE + 5.7 mm (9%) 10 (15%). best results mE always found 1.0 m, compared 1.5 m. Key factors presented methodology were: detectors customization; HoughCircle adaptability object size, distance, color; issue field natural illumination. study also highlighted uncertainty human operators reference data collection (5–6%) effect random subsampling statistical analysis estimation. Despite high values, shows potential scale. Future research will focus improving testing large scale, well investigating other image methods ability estimate growth.

Язык: Английский

Процитировано

9
Massive-Scale construction dataset synthesis through Stable Diffusion for Machine learning training DOI
Sungkook Hong, Byungjoo Choi, Youngjib Ham

и другие.

Advanced Engineering Informatics, Год журнала: 2024, Номер 62, С. 102866 - 102866

Опубликована: Окт. 1, 2024

Язык: Английский

Процитировано

6
Research and Application of YOLOv11-Based Object Segmentation in Intelligent Recognition at Construction Sites DOI Creative Commons

Luhao He,

Yongzhang Zhou, Lei Liu

и другие.

Buildings, Год журнала: 2024, Номер 14(12), С. 3777 - 3777

Опубликована: Ноя. 26, 2024

With the increasing complexity of construction site environments, robust object detection and segmentation technologies are essential for enhancing intelligent monitoring ensuring safety. This study investigates application YOLOv11-Seg, an advanced target technology, recognition on sites. The research focuses improving 13 categories, including excavators, bulldozers, cranes, workers, other equipment. methodology involves preparing a high-quality dataset through cleaning, annotation, augmentation, followed by training YOLOv11-Seg model over 351 epochs. loss function analysis indicates stable convergence, demonstrating model’s effective learning capabilities. evaluation results show [email protected] average 0.808, F1 Score(B) 0.8212, Score(M) 0.8382, with 81.56% test samples achieving confidence scores above 90%. performs effectively in static scenarios, such as equipment Xiong’an New District, dynamic real-time workers vehicles, maintaining performance even at 1080P resolution. Furthermore, it demonstrates robustness under challenging conditions, nighttime, non-construction scenes, incomplete images. concludes that exhibits strong generalization capability practical utility, providing reliable foundation safety Future work may integrate edge computing UAV to support digital transformation management.

Язык: Английский

Процитировано

4
An Integrated YOLOv5 and Hierarchical Human-Weight-First Path Planning Approach for Efficient UAV Searching Systems DOI Creative Commons
Ing-Chau Chang, Chin-En Yen,

Hao-Fu Chang

и другие.

Machines, Год журнала: 2024, Номер 12(1), С. 65 - 65

Опубликована: Янв. 16, 2024

Because the average number of missing people in our country is more than 20,000 per year, determining how to efficiently locate important. The traditional method finding involves deploying fixed cameras some hotspots capture images and using humans identify targets from these images. However, this approach, high costs are incurred sufficient order avoid blind spots, a great deal time human effort wasted identifying possible targets. Further, most AI-based search systems focus on improve body recognition model, without considering speed up shorten efficiency, which aim study. Hence, by exploiting high-mobility characteristics unmanned aerial vehicles (UAVs), study proposes an integrated YOLOv5 hierarchical human-weight-first (HWF) path planning framework serve as efficient UAV searching system, works dividing whole process into two levels. At level one, dispatched higher altitude images, covering area. Then, well-known artificial intelligence model used all persons captured compute corresponding weighted scores for each block area, according values identified bodies, clothing types, colors. two, lowers its sequentially block, descending score at it uses repeatedly until target found. Two improved algorithms, HWFR-S HWFR-D, incorporate concept convenient visit threshold weight difference, respectively, further proposed resolve issue lengthy redundant flight paths HWF. simulation results suggest that HWF, HWFR-S, HWFR-D algorithms not only effectively reduce length UAV’s blocks but also decrease required target, with much accuracy algorithms. Moreover, HWF implemented tested real scenario demonstrate capability enhancing efficiency rescue operation.

Язык: Английский

Процитировано

3
Opportunistic collaboration between heterogeneous agents using an unstructured ontology via GenAI DOI Creative Commons

Judy Akl,

Amadou Gning, Hichem Omrani

и другие.

International Journal of Intelligent Robotics and Applications, Год журнала: 2025, Номер unknown

Опубликована: Янв. 17, 2025

Язык: Английский

Процитировано

0
An Identification Algorithm of Visual Location Mark Under Ice Based on Parallel Computing DOI

Hu’ao Fan,

Zhigang Li, S. Frank Yan

и другие.

Lecture notes in electrical engineering, Год журнала: 2025, Номер unknown, С. 287 - 296

Опубликована: Янв. 1, 2025

Язык: Английский

Процитировано

0
Neural Network-Based Stress Detection in Crop Multispectral Imagery for Precision Agriculture DOI

Lídices Reyes-Hung,

Ismael Soto, A.K. Majumdar

и другие.

2022 13th International Symposium on Communication Systems, Networks and Digital Signal Processing (CSNDSP), Год журнала: 2024, Номер unknown, С. 551 - 556

Опубликована: Июль 17, 2024

Язык: Английский

Процитировано

3
Development of an automated artificial intelligence-based system for urogenital schistosomiasis diagnosis using digital image analysis techniques and a robotized microscope DOI Creative Commons
Carles Rubio Maturana, Allisson Dantas de Oliveira, Francesc Zarzuela

и другие.

PLoS neglected tropical diseases, Год журнала: 2024, Номер 18(11), С. e0012614 - e0012614

Опубликована: Ноя. 5, 2024

Background Urogenital schistosomiasis is considered a Neglected Tropical Disease (NTD) by the World Health Organization (WHO). It estimated to affect 150 million people worldwide, with high relevance in resource-poor settings of African continent. The gold-standard diagnosis still direct observation Schistosoma haematobium eggs urine samples optical microscopy. Novel diagnostic techniques based on digital image analysis Artificial Intelligence (AI) tools are suitable alternative for diagnosis. Methodology Digital images 24 sediment were acquired non-endemic settings. S . manually labeled laboratory professionals and used training YOLOv5 YOLOv8 models, which would achieve automatic detection localization eggs. Urine also employed perform binary classification detect erythrocytes/leukocytes MobileNetv3Large, EfficientNetv2, NasNetLarge models. A robotized microscope system was automatically move slide through X-Y axis auto-focus sample. Results total number 1189 labels annotated 1017 from samples. YOLOv5x demonstrated 99.3% precision, 99.4% recall, F-score, mAP0.5 detection. has an 85.6% accuracy erythrocyte/leukocyte test dataset. Convolutional neural network comparison that best options our database. Conclusions development low-cost novel identification AI be conventional microscopy This technical proof-of-principle study allows laying basis improving system, optimizing its implementation laboratories.

Язык: Английский

Процитировано

3
Validating the Use of Smart Glasses in Industrial Quality Control: A Case Study DOI Creative Commons
José Silva, Pedro Barata,

Luzia Saraiva

и другие.

Applied Sciences, Год журнала: 2024, Номер 14(5), С. 1850 - 1850

Опубликована: Фев. 23, 2024

Effective quality control is crucial in industrial manufacturing for influencing efficiency, product dependability, and customer contentment. In the constantly changing landscape of production, conventional inspection methods may fall short, prompting need inventive approaches to enhance precision productivity. this study, we investigate application smart glasses real-time during assembly processes. Our key innovation involves combining glasses’ video feed with a server-based image recognition system, utilizing advanced YOLOv8 model accurate object detection. This integration seamlessly merges mixed reality (MR) cutting-edge computer vision algorithms, offering immediate visual feedback significantly enhancing defect detection terms both speed accuracy. Carried out controlled environment, our research provides thorough evaluation system’s functionality identifies potential improvements. The findings highlight that MR elevates efficiency reliability traditional methods. synergy opens doors future advancements control, paving way more streamlined dependable ecosystems.

Язык: Английский

Процитировано

2