Analysis of Video-Based 3D Tracking Accuracy by Using Electromagnetic Tracker as a Reference
Matjaž Divjak, Damjan Zazul · 2007
When a magnetic tracker is used as a reference for vision-based tracking, a reliable transformation of their coordinate systems is crucial for proper tracking accuracy estimation. To address this issue in more detail, three different models of coordinate system alignment were developed. By analyzing the worst-case sensitivity of transformation models a limited comparison of those models is possible. The most influential model parameters are easy to identify and should be measured with special care. However, the actual parameter values used also have a significant effect on the final transformation error. With appropriate selection of parameter values, any model can be manipulated to produce the most accurate transformation. Therefore, such comparisons are only reasonable if the parameters are fixed to a certain setup of a camera and a magnetic tracker. In our experiment the transformation model A resulted in the lowest total trajectory difference, despite being the most sensitive. Statistical separation of this error into estimates of the tracking algorithm's error and the transformation-induced error provides more detailed discrepancy analysis. The presented approach is applicable to any setup where the performance of video-based tracking is to be estimated by a reference device with its separate coordinate system. Experimentally determined parameter values and conclusions are valid only for the specific setup, but the proposed methodology can be applied to any similar problem.