Fig 1.
Participants crossed an obstacle (height: 50, 150, 250 mm) at the 7th step, with their right limb being the lead limb (a). The Kinect sensors were set up in two different layouts (b). In one layout, the sensor positions (location A and location B) were located at a 20° angle and 40° angle relative to the walkway on the trail limb side; in the other layout, the sensor positions (location C and location D) were located on opposite sides, at a 75° angle and a -75° angle.
Fig 2.
A rigid body model was made for the shank segment from six infrared reflective markers. During the recording obstacle-crossing behavior, the markers on the side of the knee and ankle facing the Kinect were removed to minimize the interference between the Azure Kinect and the optical motion capture.
Fig 3.
Transformation from the Kinect’s skeleton coordinates (X′, Y′, Z′) to laboratory coordinates (X, Y, Z).
Point cloud (resolution: 640×576) data were used for the transformation.
Fig 4.
Calculation of the systematic and random error.
Residuals were calculated by subtracting Qualisys from Kinect (a). The systematic error and random error were defined as the mean and standard deviation, respectively, of the residuals (b).
Table 1.
Comparison of foot clearance measurement performance between Kinect locations.
Table 2.
Results of one-sample Wilcoxon signed-rank tests for the systematic error.
Fig 5.
Relationship between the foot clearance measured by the Kinect and Qualisys.
The diagonal straight lines represent the identity line, y = x. The black circles represent the data for the lead limb, and the white circles represent those for the trail limb. Pearson’s correlation coefficients were calculated separately for the lead and trail limbs.
Table 3.
Systematic error and random error measured in location B were compared between limbs and obstacle height conditions by using Wilcoxon signed-rank tests.