In the context of Industry 4.0, collaborative robotics can be improved by implementing appropriate reactive behavior schemes between human and the robot within the shared workspace. Inertial Measurement Units (IMUs) are widely used for real-time capturing of the human motion in clinical and sports applications, while less attention has been dedicated to IMUs for human-robot interactions in manufacturing industrial field. The aim of the present study was to validate an IMU-based method for real-time estimation of upper limbs kinematics during typical gestures required for industrial assembly tasks. To avoid ferro-magnetic disturbances typical of manufacturing environments, the magnetometers were excluded from the estimate of IMUs orientations. Shoulder and elbow angles were assessed in real-time during pick and place tasks through three IMUs fixed on the upper body of six participants. Results were validated through a stereophotogrammetric motion capture system. Errors associated to the proposed method were moderate and amounted to 2.4 and 3.5 deg on average for shoulder elevation and elbow flexion-extension angles, respectively, confirming their suitability for an industrial context of human-robot collaboration.