Development of Load Weight and Height Classifier in Lifting-Up Task Using Body Motion Metrics

Abstract

This paper presents a new weight and height class classifier for the lift-up task, focusing on future use in the control of wearable power assist devices. The proposed classifier retrieves body motion feature metrics using small accelerometers and determines the weight class (0 ky40% of the subject body weight but not exceeding 25 kg) and the height of the handle (0cm/30cm/60 cm) of the load. Three different classifiers have been trained/configured and tested. The trained classifiers are shown to correctly classify the evaluation dataset with 80-90% accuracy. An additional experiment was conducted to collect feature quantities in which subjects were requested to lift up a weight placed inside a jute bag on the floor. A discrimination accuracy of approximately 80% for the jute bag experiment verifies the versatility of the classifier for more general on-the-job lifting tasks.