Other issues with the FMC system at Coronet Peak included:
1. No exact body model existed. The local limb movements were based on the assumption that the body model was accurate and the calibration position was able to be repeated by the subject.
2. Only five sensors were used, the full body model will require at least 15 IMUs to measure the individual body segment motion.
3. The kinetics and net joint torques of the lower limbs could not be determined because skiing is predominantly dual stance.
4. It was difficult to post synchronise the GPS and IMU data.
5. Bluetooth proved unreliable for communicating with the Xbus, (IMU hub) with on average 4 out of 5 data sets corrupted.
6. There was no guarantee that the magnetic field was homogeneous over the ski area, or that the local magnetic effects of the athlete‟s equipment did not biased the results. The FMC algorithm, version one (Figure 3.4) also had some theoretical issues. Because the magnetic dip is approximately 64º, any roll error that has accumulated from the gyroscope bias will subsequently cause twice the heading error when the magnetometers are used in the correction as previously discussed in Chapter 2, page 39. The pitch and roll error were corrected first using the accelerometer channels. If there was a heading error and horizontal global acceleration was present, then the roll correction might end up causing a pitch error and instead of reducing orientation error might increase it. The solution would be to apply both corrections simultaneously, but when the static estimate of orientation was used to do this, similar to the method previously discussed (Chapter 2, page 30) the errors were much larger than the results presented here.
The prototype FMC system was introduced in Chapter 3 and for the first time it became practical to capture the motion of an athlete skiing through a complete ski run. Several issues were raised during the experiments using the prototype system. In this chapter the FMC system is developed further by addressing four important issues:
1. Validation of system performance when alternative field measurements of alpine skiing are practically difficult to obtain.
2. Development of a more robust FMC algorithm that does not rely on the global magnetic field being constant or there being no local magnetic disturbances. 3. Improvement of the accuracy of the athlete‟s body model.
4. Mapping the inertial measurement units (IMUs) to the athlete‟s body segments using a repeatable procedure.
The first two issues are addressed by the experiments in Sections 4.1 and 4.2. The prototype FMC system is compared to a video motion analysis (VMA) system in an indoor trial.
The remaining two issues are addressed by the experiments in Section 4.3, where a 3D anthropometric frame is developed and used to measure the athlete. The frame data is tested against standard anthropometric measurements of limb length and the data are used to provide the IMU to body segment calibration.
4.1. Orientation accuracy of free movement
In Chapter 2 on page 27, it was established with the pendulum experiment that IMUs are capable of accurate orientation measurements. In Chapter 3 a prototype Fusion Motion Capture (FMC) system was used to successfully track a skier at Coronet Peak but some problems with accuracy and reliability were uncovered. In this section both the vendor supplied Kalman filter and the bi-directional fusion algorithm developed in Chapter 2 for the pendulum swing, are validated against a video motion analysis (VMA) system. The orientation of a wand moving freely through space is measured using all three systems. The video analysis system is assumed to be most reliable and accurate and is used to validate the IMU based motion capture systems.
This experiment is an extension of the pendulum swing experiment. The pendulum was constrained to a single axis of movement, but the wand movements are unconstrained, they take place over long durations and the period of the movement is similar to that period of gate passing in ski racing.
4.1.1.
Method
The Wand
The orientation of a wand was tracked using an IMU and a MaxTRAQ two-camera VMA system. The T-shaped white aluminium wand had three black polystyrene balls and an IMU
attached to it (Figure 4.1). The balls were used as markers to be tracked by the VMA system as they were moved through 3D space. Because the balls were non-collinear, it was possible to obtain the orientation of the wand. The length of the wand cross bar, marker centre to marker centre was 60.4 cm, and the distance from the T-intersection to the 3rd ball was 30.1 cm. The IMU was attached at the T-intersection and its local axes were visually aligned to the local wand axes. The local axes of the wand are shown in Figure 4.2.
Figure 4.2: The wand, markers, and IMU with local axes shown
Differences between IMU and VMA estimates of orientation
To calculate the difference in measurements of orientation between the IMU and VMA measurements (θVF), the orientation measurements were first converted into quaternion form. Quaternion representation of orientation has four components, one real and three imaginary, in vector form. The imaginary parts contain the x,y, and z components of a unit vector (U) while the real part of the quaternion defines a rotation () about a vector (U, Equation 4.1).
Equation 4.1
Equation 4.2 Equation 4.3
The difference between the VMA and IMU orientation, QVF, was calculated by „quaternion
multiplication‟ of the inverse of QV by QF, (Equation 4.2). Where QV and QF are the VMA
and IMU measurements of orientation respectively. From the real element QVF(1), the
measurements were initially assumed to contain no error and so θVF was assumed to be a good
estimate of the IMU orientation error.