Location-aware sensor data error impact on autonomous crane safety monitoring

Emerging sensing technologies offer a solution to improve jobsite safety performance by providing location information to determine a worker's safety situation regarding proximity to dangers. However, due to the imperfections inherent in real-world sensor data, the collected location data might be imperfect (missing, uncertain, erroneous, and inconsistent). Among those types of imperfection, error is one of the most common. In many cases, jobsite safety monitoring applications are built on the assumption that the collected location data represent the exact situation, which might not be true due to erroneous data. However, data errors and their potential impacts on the decisions in autonomous jobsite safety monitoring systems have not been substantially studied. This paper describes an autonomous jobsite safety monitoring testbed developed to collect data from location-aware sensors. The authors developed six jobsite crane-safety monitoring test scenarios to replicate the construction activities on a full-scale jobsite and used the collected data, as well as simulated sensor data at various error levels, to quantify the impacts on decision-making performance in terms of precision and recall of workers' dangerous situations. The results indicated that the worst performance appears near the transition area of different risk levels (red to yellow/yellow to green) and the performance degrades significantly after the standard deviation of the localization errors reaches 2 cm in the testbed, corresponding to 2 m in a jobsite. The study provides researchers with an understanding of how much data errors impact safety monitoring system performance and guide future research directions.