Automatic Scaffolding Workface Assessment for Activity Analysis through Machine Learning
Round 1
Reviewer 1 Report
Summary: Authors presented a computer vision (pose estimation) (per frame) based method to identify when the construction workers are working (towards scaffolding) and when they are idle. Scaffolding being expensive, such an automated method can help save money by reducing idle time. For this they collected a video dataset. They propose to extract 3D body poses of construction workers, and learn (and test) various classifiers on top of them; targets of the classifiers being: 1) scaffold erecting, 2) transporting, 3) idle. Authors found random forests to work the best.
I like the whole idea. However, I found that somethings are missing/need to changed:
- Prepare a table containing dataset details such as total no. of samples, samples/class, etc.
- Results should be presented in a table.
- The end goal of the paper is to be able to identify idle phases. However, at least from Fig. 10, it seems that ability of classifiers to identify idle phases is low. It would be helpful and important for readers to know per class accuracy, so the authors should provide those in a table.
- Does 3D pose information yield results better than 2D pose information? If so, provide a comparison in a table. If not, then please provide an explanation why you chose to go for 3D pose information, else I would suggest to not go for 3D pose estimation, and instead go for 2D pose estimation.
- Are authors proposing 2D to 3D pose estimation used in the paper? If yes, please clearly mention it, else include the reference that proposed it.
- Pose estimation in scaffolding would be difficult, authors may want to show some failure cases?
- What are training and test sizes?
- Are authors going publicly release the dataset?
- Writing can definitely be improved. Check for typos, eg. Line 157.
This paper should be considered after authors apply above mentioned corrections.
Author Response
Please see the attached Word file.
Author Response File: Author Response.docx
Reviewer 2 Report
Please see file
Comments for author File: Comments.pdf
Author Response
Please see the attached Word file.
Author Response File: Author Response.docx
Round 2
Reviewer 2 Report
Authors should consider the below recent papers to complete the literature review related to scaffold building and 3D pose estimation.- Wang, P., Wu, P., Wang, X., Chen, X., & Zhou, T. (2020). Developing optimal scaffolding erection through the integration of lean and work posture analysis. Engineering, Construction and Architectural Management.
- Bangaru, S. S., Wang, C., Busam, S. A., & Aghazadeh, F. (2021). ANN-based automated scaffold builder activity recognition through wearable EMG and IMU sensors. Automation in Construction, 126, 103653.
- Torres Calderon, W., Roberts, D., & Golparvar-Fard, M. (2021). Synthesizing Pose Sequences from 3D Assets for Vision-Based Activity Analysis. Journal of Computing in Civil Engineering, 35(1), 04020052.
- Assadzadeh, A., Arashpour, M., Bab‐Hadiashar, A., Ngo, T., & Li, H. (2021). Automatic far‐field camera calibration for construction scene analysis. Computer‐Aided Civil and Infrastructure Engineering.
Author Response
We thank the reviewer's suggestion and all the recommended papers have been added in the revised manuscript. Please see the attachment.
Author Response File: Author Response.docx