Accelerometer-Based Activity Recognition of Workers at Construction Sites

Abstract

Although several methods have been used to capture the motion of workers at construction sites to improve productivity or safety during construction projects, collecting data by image analysis and discriminating between motions are timeconsuming processes. Therefore, it is difficult to use such methods to detect the motion of several workers or over a long time period. The authors applied an accelerometer to capture motion; this apparatus has been used in the health and medical fields. A small, light accelerometer and a simple discrimination program allowed the authors to determine whether a worker was active or inactive based on a threshold of the sum of the signal amplitude areas. Experimental surveys were conducted at construction sites: first, for setting plasterboards for two detached house construction projects and, second, for setting rebar at a large office building site. The findings of this experimental study are as follows. First, an accelerometer can be used in congested construction sites, and data can be obtained continuously. Second, this simple identification of active/inactive workers allows for the measurement of performance and the detection of problems at construction sites. Third, from the data, the hypothesis regarding workers' tendencies can be tested, that is, how workers' skill levels may affect the fluctuation of activity intensity. Therefore, in future studies, by increasing the data on workers, researchers could develop methods to improve the performance of workers.

Rebar Worker Analysis Using a 3-Axis Accelerometer

“…Mathie et al (2003) also categorized daily activities into active and inactive states and developed a program to identify them. Gondo and Miura (2020), referring to Mathie et al (2003), attached accelerometers to workers at construction sites of houses and buildings and classified them into two states: active and inactive. By doing so, Gondo and Miura (2020) were able to detect abnormal values, which coincided with the failure.…”

“…Gondo and Miura (2020), referring to Mathie et al (2003), attached accelerometers to workers at construction sites of houses and buildings and classified them into two states: active and inactive. By doing so, Gondo and Miura (2020) were able to detect abnormal values, which coincided with the failure. However, we were unable to conduct a systematic analysis of the building because of the different job types, such as rebar, carpenter among others.…”

“…In the analysis, we discriminate only three types of movements, namely, standing, walking, and sitting, to distinguish the most common movement of future workers. The present study is a continuation of Gondo and Miura (2020). While Miura's data collection took only 1 or 2 days for a specific type of work, and they encountered difficulties standardizing the content of the work, the current study records similar work, with the same type of rebar worker, over 5 days.…”

“…In addition, when a large amount of data is acquired in this way, adding to the direct classification of work and productivity analysis, there is the added possibility of revealing some worker movement characteristics by analyzing data over long periods. Based on this concern, Gondo and Miura (2020) attached a 3-axis acceleration sensor to the waist of a worker at construction sites, clarifying the degree of movement by using a simple program to distinguish between two types of workers: active and inactive. The present study aims to develop and improve the method for differentiating worker's activity, and to validate the effectiveness of data acquisition and discrimination.…”

Prediction of Labor Activity Recognition in Construction with Machine Learning Algorithms

“…Furthermore, because the construction industry is a labor-intensive sector, the impact of these technologies on labor becomes increasingly important (Calvetti et al, 2020). Therefore, much research has been done on the recognition of labor activities (Akhavian & Behzadan, 2016;Al Jassmi et al, 2021;Antwi-Afari et al, 2019;Garcia-Gonzalez et al, 2020;Gondo & Miura, 2020;Joshua & Varghese, 2014;Ryu et al, 2016;Sanhudo et al, 2021;Sherafat et al, 2020). The automatic control of laborers' work is specifically addressed by labor activity recognition processes (Ryu et al, 2019).…”