INNOVATIVE VIDEO CALIBRATION PROCEDURE USED AT ATLANTA OLYMPICS

Abstract

INTRODUCTION The objective of this study was to develop a prwedure using field dimensions and anatomical landmarks from the discus event at 1996 Atlanta Olympic Games to obtain calibration information necessary far accurate direct linear transformation (DLT) of the video records. Video records were recorded by 3 video cameras of the men s discus event at the Atlanta Olympics. Camera was located approximately 50 m behind the discus circle, camera 2 was 60 m from the right side of the circle, and camera 3 was situated 80 m away at a 45 deg to the left front of the circle in order to record the throwing scoreboard. From the rear camera view, the circle diameters next to the hash marks were digitized as control points and a scaling factor was determined using the multiplier module. Then thehash marks, circle diameter, and midline of the athlete were digitized. After scaling, the diameter of the circle was determined, and this procedure was repeated 10 times for the top 4 discus performers in the Atlanta Olympics. The data coordinate endpoints were then smoothed using a second order lowpass digital filter with a 10-Hz cutoff frequency. The average error in the 250 cm diameter dimension determined for these 40 measurements was 2.88 cm (1.2%), for a subject to camera distance of over 90 m. The position of the athlete s midline was digitized and determined using the multiplier technique from the perpendicular camera view. Next the athlete s standing height which was obtained from the Official Olympic Track and Field guide was entered into the calibration data with the x, y, coordinates of the athlete determined from the previously discussed multiplier techniques. Then using the segmental ratios reported by Dempster, the shoulder, hip, and knee heights were determined for each athlete. These heights were used to create an 3-D cube using 5 data points on the circle (left hash, left and right circle diameter, left and right sector hash) and 4 body control points. Eighteen anatomical data points, the disc, the right and left circle hash marks we1.e digitized. This composite control cube consisting of 9 points and 21 data points were digitized and entered into the DLT module and converted to real displacements. The real coordinate endpoints were smoothed using a 10 Hz cutoff frequency in a low-pass digital filter. The top 4 performers trials yielded an average error of 2.9 cm (1.2%), ranging from .6 to 3.6 cm for the 250 cm diameter from 90 m using the DLT module. CONCLUSION This multi staged approach created an 3-D cube of control points from field dimensions and anthropometric measures. This made it possible to overcome the limitation of not having a pre-determined calibration cube set in the field of view and yet, obtain accurate 3-D track and field data from the Olympics.