Extrapolation errors in controlled three-dimensional video camera set-ups

The purpose of this examination was to evaluate extrapolation errors outside the control volume in a motion analysis system by trying to control the digitising error effect that causes the inaccuracies. A Peak Performance 24 point calibration frame was used as a control object on an emulated running lane. Additionally, a second Peak Performance frame was split in eight poles, which were placed along the mid-lane at 0.5m intervals before and after the calibration frame. The positions of all 48 spherical targets were determined using a Leica 1610 electronic theodolite from three survey stations located around the measured volume. All the points were videotaped from 5 different distances with 2 cameras and digitised using the Target digitising system. The rms-deviations between the surveyed locations and the reconstructed points varied considerably in different co-ordinate directions, in different poles, and in different camera distances. The findings imply that the control points should not actually be evenly distributed around the analysed volume, but the markers should be evenly distributed in relation to cameras taking account the pan angle. In any case, extrapolation may not be as large a problem in certain situations as previously believed. There were several extrapolated markers (even up to 90% outside the calibration frame) which revealed less error than the points within the calibration.