Identify Upper Extremity Activities by Using Machine Learning

• Main Authors: LI, YU-YU, 李育瑜
• Other Authors: CHEN, HSIEH-CHING
• Format: Others
• Language: zh-TW
• Published: 2019
• Online Access: http://ndltd.ncl.edu.tw/handle/36en66

碩士 === 國立臺北科技大學 === 工業工程與管理系 === 107 === The analysis methods used to collect the actions of workers on the job site used to be conducted by video or interview. However, it is often difficult to collect and analyze such data because of the rapid and complicated movement of personnel. Therefore, the development of automated action detecting tools is a way of dealing with such difficulties and the trend in the industrial 4.0 work environment. This study simulates an assembly task in a production line and explores the applicability of using a Kinect v2 sensor with the machine learning method to detect the upper limb movements of assembly workers under both sitting and standing working postures. The experimental results show that the achieved accuracy is 0.98 in a sitting posture and 0.94 in a standing posture by using Bagging classification method. The prediction results of both sitting and standing postures can restore the sequential actions which are simulated according to predefined flow charts. This study proposes a method using Kinect to collect workers’ upper limb activities in assembly tasks and shows that a better result under a standing posture than a sitting posture. This preliminary study provides some insights for developing tools in detecting upper limb movement or in evaluating the upper limb workload of assembly workers in the future.