Fusion Motion Capture algorithm: Version four The FMC algorithms used to determine athlete motion was similar to that used for the inline

skating experiment in Chapter 5. Except, instead of gate timings, post processed and differentially corrected GPS data (pseudo-range data and carrier-frequency) were used as inputs (Figure 6.19). The snow surface model from Section 6.4 was also used to improve the

estimation of athlete vertical location. More information about the workings of the fusion algorithm was presented in Chapter 5.

Figure ‎6.19: Fusion algorithm Version Four

6.5.1.

Global trajectory

The first step was to calculate the global trajectory of the athlete‟s helmet from the attached GPS receiver and IMU. Unfortunately neither the GPS location calculated from the pseudo- range data nor the GPS location calculated from the differentially corrected integrated velocity data were accurate enough, especially in the vertical axis. Three additional constraints were required to improve the accuracy of the results to a satisfactory level:

1. The athlete remained in contact with the modelled snow surface. 2. The athlete started at the start gate and finished at the end gate. 3. The athlete passed outside each gate.

The snow contact constraint was applied in the same way as the floor contact constraint in the inline skating experiment described previously by Figure 5.5, except this time the snow slope varied according to the model (Figure 6.18).

The first estimate of the athlete‟s GPS trajectory was made by integrating the differentially corrected GPS velocity (green line, Figure 6.20). The second estimate was made by constraining the trajectory to start and finish at the boundary gates (blue line, Figure 6.20). There was a surprising amount of GPS location error based on the integrated velocity,

especially in the vertical axis. The correction required (5 metres) was around eleven times as large as the accumulated error of the walking experiment (0.4m, Figure 6.13) and more than would normally be expected for such a short duration (~16s). However, the accumulated error (around 5 metres) agrees with the expected velocity error for GPS velocity based on residuals to the carrier-frequency data during skiing (up to 0.3ms-1_{, Table 6.1).}

Figure ‎6.20: Athlete location found by integration and correction of the GPS velocity

After this correction the GPS trajectory does not appear to pass within the gate constraints (blue line, Figure 6.20). However, by fusing the GPS pseudo-range and carrier-frequency data with the IMU data (the [Fusion Algorithm 2] process in Figure 6.19) and then application of the snow surface model and gate passing constraints (the [Fusion Algorithm 3] process in Figure 6.19), a more accurate trajectory was obtained (Figure 6.21). In Figure 6.21 the athlete‟s helmet trajectory (blue line) passes outside each gate and the athlete‟s trajectory remains close to the snow surface. The gates are represented by black dots located at the bases of the inside poles.

The gate passing constraint was applied to make the athlete‟s trajectory consistent with the gate locations, which raised the following issue: The survey gate locations also contained error (up to 2 metres, Table 6.2). It was unfortunately not known which was more accurate, the athlete‟s trajectory, or the gate locations.

The gate passing constraint was applied by shifting the closest point on the athlete‟s trajectory to each gate outside that gate if required. In general a correction from skier‟s right to skier‟s left over the course was required (Figure 6.20). This solution was not ideal, but the only practical one available at the time. The resulting accuracy of the global trajectory is discussed later in section 6.9.

6.5.2.

Limb orientation and gyroscope overload

Next the individual limb segment orientations were calculated. After the head segment orientation the location of the C7 neck joint centre was obtained using the body model and processes already described in Chapter 5 on page 84. From the neck joint centre the remaining body segment orientations were obtained by working sequentially from superior body segments to inferior body segments.

However, when calculating the orientation of the thigh segment, it appeared that the thigh IMU attached to the lateral aspect of the thigh (midway between the greater trochanter and lateral femoral epicondyle) was subject to rotations that exceeded the gyroscope linear range. This was probably because of soft tissue and muscle vibrations (wobbling mass) from the ski/snow surface induced vibrations and/or a result of gate contacts.

When the gyroscope linear range (900°/s) was exceeded (at around data point 500, Figure 6.22 top panel) it resulted in a discontinuity error in the IMU orientation. The IMU orientation error can be observed by a step change in the IMU calculated global magnetic field (around data point 500, Figure 6.22 bottom panel). Over the ski course such a large step change in the measured global magnetic field is highly unlikely so it indicates an IMU orientation error was caused by gyroscope overload. The linear ranges of the gyroscopes used were 0-900°/s which was insufficient for the shock loadings that occasionally occur in skiing on some body segments. If uncorrected such a shock loading could cause transient orientation errors of up to 45 degrees in measured thigh body segment orientation.

Figure ‎6.22: Gyroscope overload (around data point 500) causes a step error

Figure ‎6.23: Global Magnetic field recalculated after the fusion process

To reduce orientation error caused by gyroscope overload the fusion algorithm checked the gyroscope channels for overload (ω>900°/s) and applied a correction during the overload periods based on assumption that both the measured global magnetic field and measured gravitational acceleration before and after the overload period were approximately equal. The validity of such a correction was determined by examining the IMU measured global magnetic field after the fusion process (Figure 6.23). The new graph shows the original step error at around data point 500 has been reduced markedly. It is surprising to observe that the measured global magnetic field is still variable over the course of the ski run. The residual

variation may be caused by local magnetic effects of the athlete‟s equipment, poor calibration of the IMU magnetometers, iron rich volcanic rock deposits on the ski area, ski area infrastructure, and/or fusion algorithm errors resulting in IMU orientation error.