Analysis and feedback of movement in manual assembly process

Abstract

A manual assembly has an important role in the manufacturing of products with small lot sizes and high variation. Becoming skilled manual labor requires knowledge transfers offered by human experts through a training process. To reduce the dependency of human experts, this paper introduces a framework called “Virtual Trainer” that incorporates the current state of the art marker-less RGB human pose estimation, activity detection for assembly step recognition, and training feedback through a multi-media presentation includes score evaluation and semantic description of trainee performance. Furthermore, the detailed transcript of each step and 3-D visualization compares to ideal movements also presented. The experimental design for evaluating the effectiveness and hypothesis is given.