Abstract
A large portion of the injuries incurred on construction sites is due to the lack of posture awareness among labors and crew while performing highly physical tasks. Most of these injuries are caused by bad and inexpert poses. Posture and gesture awareness is therefore a critical issue that can substantially decrease the number of injuries on construction sites. This paper presents a framework for identifying the level of expertise based using a machine learning-based classification algorithm. Expert and inexpert performance are identified, in order to detect unsafe and/or unproductive postures taken workers, and avoid injuries to improve productivity. The proposed framework has two major components: (1) codebook generation based on a sensorbased body joints model representing the poses taken by the labors, and (2) training a support vector machine (SVM)-based classifier for identifying expert vs inexpert performance. A set of experiments, with twenty-one masons with varying levels-of-expertise, is designed for verification and validation of the proposed methodology. An inertial measurement unit (IMU) suit, with 17 sensors, is used for data collection. While the method is applicable to any types of construction trades, this study focuses on masonry brick laying tasks. Results show that the classifier is capable of identifying the two forms with a reasonable accuracy. Moreover, SVM linear kernel classifier generates the most accurate results with lowest computational cost in classification. The low computational cost of this classifier makes it feasible for on-the-field deployment.