Login | Register

Productivity Monitoring of Construction Workers Based on Spatiotemporal Activity Recognition


Productivity Monitoring of Construction Workers Based on Spatiotemporal Activity Recognition

Torabi, Ghazaleh (2022) Productivity Monitoring of Construction Workers Based on Spatiotemporal Activity Recognition. Masters thesis, Concordia University.

[thumbnail of Torabi_MASc_S2022.pdf]
Text (application/pdf)
Torabi_MASc_S2022.pdf - Accepted Version
Available under License Spectrum Terms of Access.


Workers’ productivity monitoring is an essential but time-consuming part of large construction projects. Therefore, automating this process using surveillance cameras has gained a lot of interest among researchers. A human observer can extract both detailed and abstract information from surveillance videos to estimate productivity. Humans first gain a high-level understanding of the scene and then pay attention to low-level details. Automating this process requires computers to understand videos at different levels as well. However, previous studies only focused on low-level activities. In addition, the three-stage activity recognition method adopted by previous studies consists of separately optimized worker detection, tracking, and activity classification modules. The three-stage method propagates errors through its modules, does not leverage the scene context, and was trained and tested on trimmed datasets in the previous studies. To address these limitations and research gaps, this thesis aims to: (1) propose a fully optimized method for activity recognition of construction workers in untrimmed surveillance videos, (2) use a combination of workers’ low-level activities to understand their higher level micro-tasks, (3) calculate the percentage of workers’ time spent on different activities and micro-tasks, (4) identify low productivity and its underlying reasons by calculating the percentage of activities for each micro-task, (5) identify idling and its underlying reasons, and (6) combine resource monitoring with progress monitoring by recognizing built construction elements, calculating their completion time, the average number of utilized workers, and the percentage of their time spent on each related micro-task. The proposed fully optimized activity recognition method improved the activity classification accuracy of the three-stage method by 15%, proving that a fully optimized method is superior to the previous separately optimized methods. The proposed productivity monitoring framework was applied to a two-hour video of workers assembling footing formwork, and a six-hour video of footing formwork assembly and installation of footing reinforcement bars, showing that the framework is promising. A detailed analysis is conducted on underlying reasons for low productivity and idling, which proved that activities alone are not informative enough for decision making.

Divisions:Concordia University > Gina Cody School of Engineering and Computer Science > Electrical and Computer Engineering
Item Type:Thesis (Masters)
Authors:Torabi, Ghazaleh
Institution:Concordia University
Degree Name:M.A. Sc.
Program:Electrical and Computer Engineering
Date:3 March 2022
Thesis Supervisor(s):Bouguila, Nizar and Hammad, Amin
ID Code:990392
Deposited By: Ghazaleh Torabi
Deposited On:16 Jun 2022 15:18
Last Modified:16 Jun 2022 15:18
All items in Spectrum are protected by copyright, with all rights reserved. The use of items is governed by Spectrum's terms of access.

Repository Staff Only: item control page

Downloads per month over past year

Research related to the current document (at the CORE website)
- Research related to the current document (at the CORE website)
Back to top Back to top