Tracking Productivity in Real-time Using Computer Vision

Abstract

The construction industry is lagging behind other industries in terms of productivity gains with stagnant growth over several decades. The reasons for the lack of growth are complex and multifaceted, yet all causes and the resulting effects are realized on the output of the workers and machinery on-site. Often, typical site management practices do not identify productivity issues in a timely fashion when corrective actions may impact construction activity progress. Better management tools are required to optimize the productivity of on-site crews. This promises to positively impact the completion of individual tasks, work packages, projects, and ultimately the industry with respect to productivity, which is the driving force behind this research effort. The approach pursued focuses on utilizing commercially available cameras together with computer vision and artificial intelligence. With these tools, the main objective is to develop an algorithm capable of providing real-time feedback on the productivity of workers. Ideally, the algorithm will further identify each type of activity being conducted, thereby capturing useful productivity data. The initial model is a simple classifier that checks whether a worker performs work or stands idle. This is performed through algorithms that identify and track a person’s pose and joints and translate them to data points that can be evaluated and translated into helpful productivity measures. Finally, successfully developing a model capable of providing real-time productivity data will allow project managers and planners to better manage and utilize on-site resources. Additionally, since a large amount of data is collected and saved, trends in productivity levels can be tracked and studied further to optimize and improve them.