Estimation of Upper-Limb Orientation Based on Accelerometer and Gyroscope Measurements

Abstract

A solution is proposed to the estimation of upper-limb orientation using miniature accelerometers and gyroscopes. This type of measurement device has many different possible applications, ranging from clinical use with patients presenting a number of conditions such as upper motor neuron syndrome and pathologies that give rise to loss of dexterity, to competitive sports training and virtual reality. Here we focus on a design that minimizes the number of sensors whilst delivering estimates of known accuracy over a defined frequency range. Minimizing the sensor count can make the measurement system less obtrusive, as well as minimising cost and reducing the required bandwidth if using a wireless solution. Accurate measurement of movement amplitude up to 15 Hz is required in our immediate application, namely to quantify tremor in multiple sclerosis patients. The drive for low numbers of sensors and good accuracy at higher frequencies leads to a novel design based on composite filters. The simple estimator structure also gives good insight into the fundamental accuracy limitations based on the sensors chosen. This paper defines the underlying mathematics, and quantifies performance for an estimator for shoulder, upper arm, lower arm and hand orientations. Good estimation accuracy up to 15 Hz is indicated, and this with a reduced total sensor count of 18 compared to 24 that would be required for more conventional estimator architectures.

Automatic Activity Classification and Movement Assessment During a Sports Training Session Using Wearable Inertial Sensors

“…However, it has some disadvantages including implementation complexity [26], [27], high sampling rate due to linear regression iteration (fundamental to the Kalman process) and the requirement to deal with large scale vectors to describe rotational kinematics in three-dimensions [25], [10]. There are some other alternatives to address these issues including Fuzzy processing [28] or frequency domain filters [29]. Although these approaches are easy to implement, they are limited to operating conditions.…”

Toward Automatic Activity Classification and Movement Assessment During a Sports Training Session

“…However, it has some disadvantages including implementation complexity [39], [40], high sampling rate due to linear regression iteration (fundamental to the Kalman process) and the requirement to deal with large scale vectors to describe rotational kinematics in three-dimensions [38], [16]. There are some other alternatives to address these issues including Fuzzy processing [41] or frequency domain filters [42]. Although these approaches are easy to implement, they are limited to operating conditions.…”

Measuring and modelling arm dynamics to support studies into reducing tremor in individuals with multiple sclerosis

“…The method described by Hyde et al [7], where the body segments are treated as a rigid body system joined together at joints with specified degrees of freedom (DOF), will be used to enhance the accuracy of estimates.…”

“…The SimMechanics environment has the added advantage that it is easier to use frequency domain techniques, such as composite filters, to combine measurement device readings, as described by Hyde et al [7]. In this way, the most appropriate readings from the accelerometers, gyros and magnetometers contribute to the estimate of the body movement in the frequency range to which they are best suited.…”