Fingertip Force and Muscle Activation Patterns at Varying grasp Objects

• 영문명: Fingertip Force and Muscle Activation Patterns at Varying grasp Objects
• 발행기관: 호서대학교 기초과학연구소
• 저자명: Suji Park Juhyun Park Seyeon Oh Chaeyeon Heo Sieun Ho Seonhong Hwang
• 간행물 정보: 『기초과학연구 논문집』제30권 제1호, 32~51쪽, 전체 20쪽
• 주제분류: 공학 > 공학일반
• 파일형태: PDF
• 발행일자: 2022.12.31

영문 초록

In this study, we tried to collect and analyze the kinetic and neurological information such as finger-tip forces and EMG for several representative (the most commonly used) grasp movements to explore their force and muscle activation patterns based on the newly defined grasp taxonomy. Ten able-bodied (five males, five females) volunteered to participate and they performed five different grasp tasks: holding a bottle (Bottle), turning a doorknob (Knob), cutting with a knife (Knife), brushing with a toothbrush (Toothbrush), holding a thick book (Book) after we attached five force sensitive resistor (FSR) sensors on the tip of fingers and four surface electromyogram (sEMG) electrodes on the lower arm of the subject’s dominant hand. Root Mean Square (RMS) and Mean Absolute Value (MAV) from the mean maximum values of sEMG(%) and fingertip force(kgf) of all ten subjects were extracted as features. The classification from the feature dataset using convolutional neural network (CNN) was applied and analyzed the results of accuracy and repeatability. The mean maximum values of EMG and fingertip forces during five different grasp tasks, and the MAV and RMS which were extracted features from the above were compared with task pairs. They showed significant differences in comparison of four pairs of tasks which were Bottle and Knife (p = 0.005 in both MAV and RMS), Bottle and Toothbrush (p = 0.005in both MAV and RMS), Bottle and Book (p = 0.013 in both MAV and RMS), Knob and Toothbrush (p = 0.047 in MAV and p = 0.028 in RMS). The classification accuracy of the Bottle grasp task was the largest at 60% (true positive predictive rate is 60% and false postive rate is 40%), while the other tasks showed an 30-40% of accuracy. Repeatability was 60% in the Bottle task and 50% in the Knob task, and those of the other tasks were ranged 30-40%. Overall, it is believed that the small number of samples in the study is the main reason of the low accuracy and repeatability of the classification. A total of nine variables (four sEMG and five forces) showed different significances in paired mean comparisons for five grasp tasks (graspping a bottle, turning a doorknob, cutting with a knife, brushing teeth with a toothbrush, holding a thick book). A comparison of the reduced variable from feature extraction also showed different classification accuracy for five grasp tasks.

목차

Ⅰ. Introduction

Ⅱ. Methods

Ⅲ. Results

Ⅳ. Discussion

Ⅴ. Conclusion

Ⅵ. References

키워드

Hand grasp Fingertip force Electromyography Force sensistive resistor