Transcript
A COMPARISON OF POSITION MEASUREMENT ACCURACY USING TWO DIFFERENT CAMERA ARRANGEMENTS W. Matt Denning, Iain Hunter, and Matthew K. Seeley Brigham Young University, Provo, UT, USA Email:
[email protected] INTRODUCTION Motion analysis, using high-speed video and reflective markers placed on anatomical landmarks, is an important measurement method in biomechanical research. The accurate measurement of reflective marker position during motion is vital for various biomechanical analyses. Researchers have studied the influence of marker placement and size, and calibration technique on the accuracy of marker position measurement [1, 2]; however, it is unclear if/how the distance between the video cameras and reflective markers influence marker position measurement accuracy. We currently use two primary camera arrangements in our lab. One is an arrangement that uses cameras placed on the floor, relatively close to the subjects (~2 m). The other arrangement involves cameras that are mounted on the wall at a relatively far distance from the subjects (~4 m). The purpose of this study was to determine whether one of these two camera arrangements results in a more accurate measurement of reflective marker position. We hypothesized that an increased distance between cameras and reflective markers would result in decreased marker position measurement accuracy. METHODS Spatial position was measured for four 13-mm static reflective markers. These markers were placed in a known position and the distance between each marker was calculated using the measured (using high-speed video) and known positions. We also measured position for one dynamic reflective marker after it had been dropped into the motion capture volume from a height of 1.5-m. Position data for three trials were averaged for both the static and dynamic reflective markers. For the dynamic marker, vertical acceleration was derived using a central differences approach. All of these
measurements were made using two camera arrangements: (1) a relatively close arrangement, with cameras positioned 2 m from a set global origin, and (2) a relatively far arrangement, with cameras positioned 4 m from a set global origin (Figure 1). Six MX13T cameras, two MXF20 cameras, and two MXT20 cameras, and VICON Nexus software (VICON, Santa Monica, CA, USA) were used to collect the position data (240 Hz). A wand calibration was completed using 2000 images spaced throughout a 1 × 1 × 1m volume. A calibration error of less than 0.2 was accepted as a successful calibration for each camera, as is recommended by the manufacturer.
Figure 1: A representation of the two camera arrangements. Dark and light trapezoids represent the close and far camera arrangements, respectively. Small black spheres represent the static markers. The independent variable was camera arrangement (close and far). Dependent variables were: (1) root mean square error (RMSE: the difference between the known and measured distances (between the static markers)), and (2) vertical acceleration of the dynamic marker within and without the calibrated motion capture volume. We used a one tailed t-test to test the effect of the independent variable on the dependent variables (α = 0.05).
Acceleration y (m/s/s)
RESULTS AND DISCUSSION The present results did not support our hypothesis. Mean RMSE for the relatively close position (0.28 mm) was not statistically different from mean RMSE for the relatively far position (0.49 mm; p = 0.12; Figure 2). Similarly, no significant difference in vertical acceleration was noted between the close and far camera arrangements (p = 0.44; Table 1). There is no statistical difference in position measurement accuracy between close and far camera arrangements.
RMSE (mm)
0.25
0.5 Time (s)
0.75
0.00 3 to 4
Figure 2: Difference in marker distance (relative to the known distance) for two camera arrangements.
In conclusion, the purpose of this project was to determine whether camera distance affects position measurement accuracy. Statistically, position measurement accuracy was similar for close and far camera arrangements. Additionally, accuracy is greatest when the measurement occurs inside the calibrated volume. Future research could strengthen these findings by using additional markers to further investigate the influence of camera placement on position and acceleration accuracy.
Mean vertical acceleration for the dynamic marker, for the close and far camera arrangements, within and without the calibration volume, is shown in Figure 3. Although this was not the primary purpose of the present study and a statistical analysis was not performed, vertical acceleration for the relatively far camera arrangement appear to have been more variable (Figure 3 and Table 1). This implies that the position data were more variable for the far camera arrangement. Moreover, standard deviations outside the calibration volume are much higher for the far camera arrangement (Table 1), supporting the idea that it is important to measure position within the calibrated volume, especially
REFERENCES 1. Windolf M, et al. J Biomech 41, 2776-2780, 2004. 2. Telfer S, et al. Gait & Post 32, 536-539, 2010.
Table 1: Average accelerations (m/s/s) within and without the calibration volume. Without Mean SD
Close -9.86 4.42
1
when accelerations need to be calculated. Additionally, as noise increases when deriving acceleration from position data, perhaps a different smoothing technique should be used to eliminate noise from the true signal, when accelerations are calculated outside the calibration volume.
0.30
1 to 4 2 to 3 2 to 4 Measured Distances
-15
Figure 3: Vertical acceleration for the dynamic marker, plotted against time, for the two camera arrangements. Acceleration to the left of the vertical dotted line occurred without the calibration volume. It appears that vertical acceleration was more stable when measured in the volume.
Far
1 to 3
-10
0
Close
1 to 2
-5
-20
0.90
0.60
Close Far
0
Within Far -10.0 6.29
Close -9.48 0.92
Far -9.52 1.92
