On construction sites, workers are exposed to a wide range of hazards. To reduce risk and prevent severe work accidents, the use of PPE (Personal Protective Equipment), such as hard hat or harness has been adopted by the industry. Monitoring the use of PPE is an inherently difficult task since its protective effectiveness depends on whether it is conformally worn. In recent years, the advent of deep learning for the tasks of object detection and human pose analysis has enabled the rise of discriminative and all-purpose PPE detectors. However, most of earlier works solely focus on detecting the equipment irrespective of their positioning. By assessing conformal PPE usage based on the combined output of object detection and articulated human pose extraction we can combine learning with domain expertise, which allows a straightforward, in-field application of the approach that attains high quantitative performance.