Analysis of Video-Based 3D Tracking Accuracy by Using Electromagnetic Tracker as a Reference

Abstract

In recent years video-based tracking systems have been gaining widespread attention in several application fields. They are often used in military or surveillance applications (Ellis & Black, 2003; Collins et al, 2000; Cupillard et al., 2003; Fischer et al.,2004; Safeguards, 2007), in medicine (Grimson et al., 1998; Bornik et al, 2003; Pandya & Siadat, 2001; Tang et al., 1998; Bernd & Seibert, 2004), entertainment industry (Stapleton et al., 2002; Wren et al., 1997; Huang & Yan, 2002; Collomosse et al., 2003; Fua & Plankers, 2003) or sport (Qiu et al., 2004; Gueziec, 2002; Kristan et al., 2006), for research on human-computer interaction (Sato et al., 2004; Bradley & Roth, 2005; Polat et al., 2003), intelligent environments (Krumm et al., 2000) and similar. Continuous technological development and increasing competition among vendors have led to a great selection of tracking systems that are available on the market today with a variety of capabilities. To compare them, several factors have to be considered. While price, speed or technical limitations may be very important for initial selection, the tracking accuracy is usually the most important property. To assess a tracking system and its precision, we need a reliable measure which allows for comparison of tracking system performance, provides estimates of tracking errors and indicates how to optimize the tracking system parameters. The natural way to analyze the accuracy of any tracking system is to compare it to some reliable reference data. While a selection of comparison methods is readily available to the research community (Needham & Boyle, 2003), a reliable reference data (ground truth) can be hard to obtain, especially if greater accuracy is desired. Publicly available collections of video recordings with registered 3D ground truth information can be helpful, but are very scarce and with limited selection (Scharstein & Szeliski, 2003; CVTI, 2007). Such collections can be very useful in the development and testing of tracking algorithms, but are not enough for evaluation of a complex video tracking system in its actual operating environment. Instead, one of the most popular approaches to obtain the reference data is to resort to an electromagnetic tracking device. These devices offer fast and accurate measurements and are insensitive to the line-of-sight requirements of optical motion trackers, which makes them ideally suited for tracking free-moving objects.