Construction worker productivity evaluation using action recognition for foreign labor training and education: A case study of Taiwan

Importing workers from countries with more abundant, lower-cost labor supplies has been a solution in developed countries to address the labor shortages. The continuous monitoring of worker activities is a widely used approach to measuring productivity and determining the root causes of inefficiency. The evaluation of column formwork activities was developed and tested to compare the relative productivity of foreign and domestic workers. An innovative framework integrating a deep learning method, namely You Only Watched Once (YOWO), was applied to recognize discrete actions of the workers. The YOWO model identified these actions with an average F1 score of 78.3%. The time required to perform the "assemble" action was identified as the most significantly different, with foreign workers requiring 36 s more than their domestic counterparts. Hence improving the efficacy of foreign workers in completing the assemble action should be prioritized in training and education.