Infrared marker tracking cameras occupy a significant amount of space in the operating theatre, and require a constant line-of-sight between camera and markers which reduces patient access. We therefore investigate the accuracy of a novel, single, drill-mounted, commercial web camera using augmented reality. The system is built upon the ARToolKit library and provides full six degrees of freedom tracking of the tool relative to fiducial markers. Tool positioning accuracy was assessed using three methods. Firstly, the camera was displaced linearly along each orthogonal axis, relative to a marker, in 1 mm intervals over a range of 150 mm. Secondly, a 100×100×50 mm pyramidal target with regular measurement points was machined to an accuracy of 10 μm. 108 points were probed with the system producing 100 measurements for each. These were performed with the camera both static and randomly rotated during measurement. Finally, the probe was systematically traced across the surface of the pyramidal target for a period of 5 minutes, resulting in approximately 10,000 positional measurements. Linear displacements produced RMS precision errors of 1.4 mm along the optical axis at separations above 250 mm, however, these errors reduced to 0.4 mm for separations below 180 mm. Axes orthogonal to the optical axis produced RMS errors of 0.3 mm at approximately 200 mm separation. The point experiment produced a total RMS accuracy error of 1.5 mm while the surface trace experiment produced a total RMS error of 1.7 mm. The results demonstrate two commonly reported features of existing optical tracking systems. Namely, system accuracy is inversely proportional to camera-marker separation and the optical axis typically presents the lowest accuracy. The drill mounted camera approach capitalises upon this first effect by allowing substantially reduced camera-marker separation, compared to existing systems, particularly during resection. Without published tool accuracies for existing systems it is difficult to confidently define a success threshold, and with features such as overcutting to facilitate implant cementation the situation is further complicated. However, it is reasonable to suggest that submillimetre accuracies are required for consistently successful arthroplasty. The results currently indicate that the system falls short of this threshold. However, several optimisation techniques have yet to be implemented, including improved camera calibration and increased image resolution. In conclusion, one-camera augmented reality systems may have the potential to replace the current optical pathway. As such, future work will focus on optimising the system to reach the desired level of accuracy.
