This research on identification and classification of construction workers' activity contributes to the monitoring and management of individuals. Since a single sensor cannot meet management requirements of a complex construction environment, and integrated multiple sensors usually lack systemic flexibility and stability, this paper proposes an approach to construction-activity recognition based on smartphones. The accelerometers and gyroscopes embedded in smartphones were utilized to collect three-axis acceleration and angle data of eight main activities with relatively high frequency in simulated floor-reinforcing steel work. Data acquisition from multiple body parts enhanced the dimensionality of activity features to better distinguish between different activities. The CART algorithm of a decision tree was adopted to build a classification training model whose effectiveness was evaluated and verified through cross-validation. The results showed that the accuracy of classification for overall samples was up to 89.85% and the accuracy of prediction was 94.91%. The feasibility of using smartphones as data-acquisition tools in construction management was verified. Moreover, it was proved that the combination of a decision-tree algorithm with smartphones could achieve complex activity classification and identification.
Keywords: activity recognition; construction management; feature extraction; machine learning; sensor; smartphone.