Temporal image analytics for abnormal construction activity identification

Automation in Construction, Journal Year: 2021, Volume and Issue: 124, P. 103572 - 103572

Published: Jan. 27, 2021

Language: Английский

---

Measuring and benchmarking the productivity of excavators in infrastructure projects: A deep neural network approach

Automation in Construction, Journal Year: 2021, Volume and Issue: 124, P. 103532 - 103532

Published: Jan. 27, 2021

Language: Английский

---

A semi-supervised learning detection method for vision-based monitoring of construction sites by integrating teacher-student networks and data augmentation

Advanced Engineering Informatics, Journal Year: 2021, Volume and Issue: 50, P. 101372 - 101372

Published: Aug. 11, 2021

Recently, deep-learning detection methods have achieved huge success in the vision-based monitoring of construction sites terms safety control and productivity analysis. However, require large-scale datasets for training purposes, such are difficult to develop due limited accessibility images need labor-intensive annotations. To address this problem, research proposes a semi-supervised learning method site based on teacher–student networks data augmentation. The proposed requires number labeled achieve high performance scenarios. Initially, trains teacher object detector with following weak Next, trained generates pseudo-detection results from unlabeled that been weakly augmented. Finally, student is both strongly In our experiments, 10,000 annotated Alberta Construction Image Dataset (ACID) divided into set (70%) validation (30%). 91% mean average precision (mAP) while only requiring 30% set. comparison, existing supervised ResNet50 Faster R-CNN mAP 90.8% when full These experimental show potential reducing time, effort, costs spent developing datasets. As such, has explored increased practicality systems industry.

Language: Английский

---

3D convolutional neural network‐based one‐stage model for real‐time action detection in video of construction equipment

Computer-Aided Civil and Infrastructure Engineering, Journal Year: 2021, Volume and Issue: 37(1), P. 126 - 142

Published: June 10, 2021

Abstract This study aims to propose a three‐dimensional convolutional neural network (3D CNN)‐based one‐stage model for real‐time action detection in video of construction equipment (ADVICE). The 3D CNN‐based single‐stream feature extraction and are designed with the implementation attention module pyramid developed this improve performance. For evaluation, 130 videos were collected from YouTube including four types at various sites. Trained on 520 clips tested 260 clips, ADVICE achieved precision recall 82.1% 83.1%, respectively, an inference speed 36.6 frames per second. evaluation results indicate that proposed method can implement diverse, variable, complex paved way improving safety, productivity, environmental management projects.

Language: Английский

---

Recognition of pedestrian trajectories and attributes with computer vision and deep learning techniques

Advanced Engineering Informatics, Journal Year: 2021, Volume and Issue: 49, P. 101356 - 101356

Published: July 13, 2021

Language: Английский

---