Evaluating the effectiveness of training image dataset for computer vision applications in construction
Automation in civil engineering – and more specifically, in the construction industry – has long been regarded as the next phase in its evolution with the times. However, a resistance to harness new technologies has seen the industry fall behind many of its counterparts more willing to embrace the l...
Saved in:
| Main Author: | |
|---|---|
| Other Authors: | |
| Format: | Final Year Project |
| Language: | English |
| Published: |
Nanyang Technological University
2024
|
| Subjects: | |
| Online Access: | https://hdl.handle.net/10356/177288 |
| Tags: |
Add Tag
No Tags, Be the first to tag this record!
|
| Institution: | Nanyang Technological University |
| Language: | English |
| Summary: | Automation in civil engineering – and more specifically, in the construction industry – has long been regarded as the next phase in its evolution with the times. However, a resistance to harness new technologies has seen the industry fall behind many of its counterparts more willing to embrace the latest and greatest. One critical aspect of this evolution involves equipping construction site machineries with the ability to understand visual scenes. The identification of human workers is seen as a fundamental step towards achieving the end goal of complete automation. In order to accomplish this, object detection algorithms can be implemented.
This study aims to compare and identify the best performing object detection model in the context of worker identification in a construction setting. Furthermore, this study also explores the importance and effectiveness of the quantity of training data. In a field where training data is scarce and often found wanting, it is important to evaluate the most optimal training dataset sizes, and consequently investigate the significance and impact of insufficiently large training data sets. |
|---|