Abstract —One of the biggest challenges facing Americans working in industrial factories is the risk of developing musculoskeletal disorders (MSDs). About 2% of all American workers suffer from MSDs every year. This has a significant social consequence on the lives of workers and places a large burden on the employers, as MSDs account for over one- third of all worker compensation costs. Still-frame models developed to reduce work-related MSDs either require expertise or lack real-time analysis. The focus of this project was to design an accurate monitoring system that could help factory workers correct their heavy-lifting technique by making adaptive technique recommendations. To observe human lifts, we made use of the Microsoft Kinect depth sensing camera, which has the ability to provide real-time skeletal tracking at 30 frames per second. Proper lifting techniques were defined using several lifting equations and various biomechanical models. Knowledge of the user’s joint angles allows us to assess lift safety. In our system design, users are first asked to perform several lifts in different canonical styles. The system then provides the user with a recommended lifting style that maximizes safety within the constraints of the user’s measured capabilities. I. INTRODUCTION Musculoskeletal Disorders (MSDs) affect nerves, tissues, joints and supporting structures, such as intervertebral discs. About two percent of all American workers suffer from work-related musculoskeletal diseases (WMSDs) every year [1]. This accounts for over one-third of all worker compensation costs [1]. Annual direct and indirect health costs for MSDs were $950 billion in 2004, which was equal to 7.4% of the gross domestic product [2]. This enormous cost of MSDs and other work related injuries have prompted companies to explore ways of reducing the risk of MSDs in factory workers. Manuscript received April 1, 2013. P. A. Beling is an Associate Professor in the Department of Systems and Information Engineering at the University of Virginia. Jeffrey Delpresto (email: jvd9ju@virginia.edu), Chuhong Duan (email: cd8dz@virginia.edu), Lara M. Layiktez (email:lml6hr@virginia.edu), Eyitemi G. Moju-Igbene (email:egm7ch@virginia.edu) Matthew B. Wood (email:mbw3td@virginia.edu) are undergraduate students of Systems Engineering at the University of Virginia. MSDs result mainly from physically-demanding tasks at workplaces [3]. Most adults spend a substantial portion of their day at work. Thus, occupational safety hazards can easily affect many aspects of their lives, especially their health. The purpose of this project was to develop a system that would decrease the risk of injury for factory workers while lifting heavy objects. Workers are often unaware of the risk they place upon themselves when lifting improperly. This paper is based on the hypothesis that worker safety can be improved through education on proper lifting technique. We propose an adaptive training program that provides the user with a recommended lifting style that maximizes safety within the constraints of the user’s measured capabilities. The system uses the Microsoft Kinect to collect a user’s skeletal data, and analyzes this data, using Microsoft Software Development Kit (SDK), to determine if the user’s lifting technique is safe. Users perform two of the major lifting techniques, a one knee kneeling lift, and a two knee squatting lift [4]. A qualitative model, based on seven lifting criteria is used to characterize proper lifting technique. The system then recommends a technique to the user (kneeling or squatting) based on users’ flexibility and ability to perform both lifts. A. Alternative Systems Alternate systems include the motion capture systems created by the Motion Analysis Corporation and Vicon Motion [5], [6]. Both systems comprise a hardware monitoring component and a software analysis component. The monitoring hardware used in both systems differs from the Microsoft Kinect sensor in that they collect data in the optical light spectrum whereas Kinect receives infrared data. They require users to wear physical sensors called “markers.” These allow for a variable number of joints to be analyzed and a 3D biomechanical assessment of human interaction with products, workspaces or living spaces. The Microsoft Kinect has the improved capability to analyze user’s movements without “markers.” However, it loses some precision when interacting with objects. The software components of both Motion Analysis Corporation and Vicon systems can interface with the Siemens Corporation’s JACK, a human modeling software package. JACK uses quantitative lifting data to provide the user with feedback on their lifting technique [7]. This differs from our system that uses qualitative models in order to determine the appropriate lifting recommendation. Safe Lifting: An Adaptive Training System for Factory Workers Using the Microsoft Kinect Jeffrey Delpresto, Chuhong Duan, Lara M. Layiktez, Eyitemi G. Moju-Igbene, Matthew B. Wood, and Peter A. Beling Member, IEEE