Improving Data Glove Accuracy and Usability Using a Neural Network When Measuring Finger Joint Range of Motion

Abstract

Data gloves capable of measuring finger joint kinematics can provide objective range of motion information useful for clinical hand assessment and rehabilitation. Data glove sensors are strategically placed over specific finger joints to detect movement of the wearers’ hand. The construction of the sensors used in a data glove, the number of sensors used, and their positioning on each finger joint are influenced by the intended use case. Although most glove sensors provide reasonably stable linear output, this stability is influenced externally by the physical structure of the data glove sensors, as well as the wearer’s hand size relative to the data glove, and the elastic nature of materials used in its construction. Data gloves typically require a complex calibration method before use. Calibration may not be possible when wearers have disabled hands or limited joint flexibility, and so limits those who can use a data glove within a clinical context. This paper examines and describes a unique approach to calibration and angular calculation using a neural network that improves data glove repeatability and accuracy measurements without the requirement for data glove calibration. Results demonstrate an overall improvement in data glove measurements. This is particularly relevant when the data glove is used with those who have limited joint mobility and cannot physically complete data glove calibration.