Human activity recognition and mobile sensing for construction simulation

Abstract

Construction industry has been constantly lagging behind in terms of efficiency and productivity growth. Simulation modeling can be used to improve the productivity of construction workflow processes through modeling uncertainties and stochastic events that may negatively impact project cost and schedule. In the research presented in this paper, mobile sensors coupled with machine learning techniques are used for ubiquitous data collection and human activity recognition (HAR), which will constitute the key input parameters of process simulation modeling. To assess the designed methodology, an experiment is carried out which replicates a warehouse quality control operation. Smartphones mounted on human bodies are used to collect multi-modal time-motion data. Support vector machine (SVM) is then applied to classify workers' and inspectors' activities, and activity durations are subsequently extracted. Finally, a simulation model is built using the output of the HAR phase and rigorously validated and used to analyze workflow processes, productivity, and bottlenecks.