Please use this identifier to cite or link to this item: https://scholarbank.nus.edu.sg/handle/10635/222247
Title: EVALUATING THE PERFORMANCES OF VARIOUS DEEP LEARNING ALGORITHMS IN LIFTED LOAD DETECTION FOR CONSTRUCTION SAFETY
Authors: TUI YUN NING ADELINE
Keywords: Building
PFM
Project and Facilities Management
Goh Yang Miang
2019/2020 PFM
Computer Vision
Deep Learning
Faster R-CNN
SSD
YOLO
Load detection
Workplace safety and health
Issue Date: 2019
Citation: TUI YUN NING ADELINE (2019). EVALUATING THE PERFORMANCES OF VARIOUS DEEP LEARNING ALGORITHMS IN LIFTED LOAD DETECTION FOR CONSTRUCTION SAFETY. ScholarBank@NUS Repository.
Abstract: Construction workers operate in a highly intricate environment, exposed to a host of inherent risks including fall from height, struck by object, electrocution and caught- in/between. Struck by Falling Object (SBFO) is the second leading cause of all workplace injuries and fatalities on construction and demolition sites. More often than not, workers suffer bruises due to falling loads from cranes or other hoisting equipment. To address this alarming issue, government bodies and construction companies worldwide have developed and implemented various preventive measures. These measures, however, are deemed to be inadequate as construction-related injuries and fatalities remained unacceptably high, lagging behind other major industries. Computer Vision (CV) is increasingly popular for vision-based monitoring to improve construction safety and health on-site. Building on the current state-of-the-art of CV, this research aims to evaluate the effectiveness of various Deep Learning algorithms in lifted load detection. To achieve this aim, three experiments namely YOLOv3, Faster R-CNN and SSD were conducted. The general workflow in conducting these experiments include gathering of video data from the Singapore Housing and Development Board, screening of videos, extracting of key video frames, labelling of images, training and testing of model. All the trained models were evaluated according to the protocol of PASCAL VOC detection challenge 2010-2012. Based on the evaluated results, the YOLOv3-2 model provided a higher mean Average Precision for majority of the load classes namely Precast plank, Precast Column and Household Shelter. Therefore, this study provides a foundation for future research in enhancing the applicability of the YOLOv3, Faster R- CNN and SSD model to real construction sites. This helps to minimize the number of workplace related injuries and fatalities due to contact with objects.
URI: https://scholarbank.nus.edu.sg/handle/10635/222247
Appears in Collections:Bachelor's Theses

Show full item record
Files in This Item:
File Description SizeFormatAccess SettingsVersion 
Tui Yun Ning Adeline 2019-2020.pdf26.42 MBAdobe PDF

RESTRICTED

NoneLog In

Page view(s)

41
checked on Sep 29, 2022

Download(s)

19
checked on Sep 29, 2022

Google ScholarTM

Check


Items in DSpace are protected by copyright, with all rights reserved, unless otherwise indicated.