ACCURACY OF DIGITIZATION USING AUTOMATED AND MANUAL METHODS

Abstract

The objective of this study was to determine the limits of accuracy of angular estimates produced by a computerized three-dimensional (3-D) motion measurement system using auto-digitization and manual digitization. We hypothesized that auto-digitization would produce more accurate and reliable estimates of angular data than manual digitization. Twelve reference angles were estimated from markers placed on a T-shaped pendulum. The pendulum’s angular velocity was systematically increased by changing the release position from four angles (0 -static, 45 , 90 , and 120 ). Angular estimates were compared over 20 frames for 10 trials at each release position using analysis of variance (ANOVA), and intra-class correlation coefficients (ICC’s) were used to estimate the variability among release positions. Two S-VHS camcorders were placed at 45 angles relative to the activity plane for filming, forming an inter-camera angle of 90 . The cameras operated at a film speed of 60 Hz, with a shutter speed of 1/500 second. A single 300-W flood light was positioned beside each camera to illuminate 1.90 cm diameter retro reflective markers. A 3-D calibration structure defined a calibration area 182.88 cm wide, 152.40 cm high, and 60.96 cm deep. Eighty independent film clips (4 angular velocities X 10 trials X 2 views) were manually digitized across 20 frames by five experienced digitizers. Data was transformed to 3-D coordinates using a DLT, and smoothed with a Butterworth second-order, low-pass recursive filter with a fc of 6 Hz. Mean errors of the angular estimates across trials and frames were within ±1.0 for each of the four release positions. ANOVA and a post-hoc Tukey test revealed the mean error of the auto-digitized trials was statistically larger (p<0.0001) than the manually digitized trials. For the auto-digitized trials, the static (release position 0 ) produced statistically less (p<0.05) mean error than the dynamic trials. The ICC’s showed a high degree of consistency among all digitizers, with the range from 0.707 to 0.999. The results of this study support the conclusion than clinically acceptable accuracy can be obtained regardless of digitization method across a range of angular velocities. Surprisingly, it revealed higher accuracy for manually digitized angular estimates than for auto-digitized estimates.