Carlomagno D ¹, Sassi R ², Bosio A ¹, Connolly DR ¹, Rampinini E^1,3

1, Human Performance Laboratory, MAPEI Sport Research Center, Castellanza, Varese, ITALY;2, FC Juventus, Turin, ITALY;3, Department of Sport, Nutrition and Health Sciences, Faculty of Exercise and Sports Sciences, Università degli Studi di Milano, Milan, ITALY

Commercial semi-automated, video-computerized, image recognition match analysis systems can now supply soccer teams with match analysis data on all players. Despite the popularity of these systems among professional soccer teams and researchers, there have been few studies investigating the validity of these systems to measure distances at different intensities. The aim of this study was to provide preliminary evidence on the accuracy of Sport VU (Tel Aviv, Israel), a commercially available video match analysis system. We also examined systems accuracy in calculating the mean metabolic power (P_mean) sustained by the athletes. Six soccer players performed 48 bouts of intense match simulation exercise including sprinting, striding and jogging in 4 different zones of the pitch at different distance and angles from the video system. Player velocities measured using the video-analysis system were compared to those registered by a radar system (Stalker, ATS, Texas, USA). The data collected by the radar system was filtered using a zero lag Butterworth filter, and the first and last 10% of the data were excluded from analysis in order to avoid the edge effect due to the filtering algorithm. Accuracy of high-intensity running (HIR >15km/h), very high-intensity running (VHIR >20km/h), total distance (TD) and the P_mean of the intense match simulations were determined using the typical error (TE) expressed as a coefficient of variation. The systematic bias was also verified. Overall, a small TE was observed for TD (0.8%), HIR (2.1%) and VHIR (2.5%). While TE for P_mean was 5.8%. It was noted that TE tended to vary between the 4 different zones of the pitch, ranging from 0.3–1.2%, 1.1–3.3%, 1.9–3.2%, 2.4–8.8% for TD, HIR, VHIR and P_mean respectively. The mean systematic bias for TD, HIR, VHIR and P_mean were -1.7%, -0.8%, 0.4% and 32.3% respectively. The preliminary results of this study report a high level of accuracy for distances covered and the intensity of running performed by the soccer players. However P_mean values calculated from the video data should be interpreted with caution due to the large over-estimation observed. Studies with larger sample sizes and using 90 minute soccer match simulation are necessary to fully validate video-match analysis systems. In addition, errors in the accuracy of the system due to positional changes with respect to the camera need to be further investigated.

The 3rd World Conference on Science and Soccer in Ghent, Belgium: 14-16 May 2012