The average and variance of the reprojection error is shown in Table 3, together with the groundtruth checkerboard’s reprojection error. The peaks in the reprojection error from the fingertip pose estimation can be explained by abrupt inaccurate locations of fingertips. Since
the five point correspondences from the fingertips are close
to the minimum number for the camera pose estimation algorithm, small deviations can cause error spikes in the results.
In order to make the system more robust, occasional jitters are filtered out by applying our Kalman filter with fine-tuned noise covariance matrices and thresholds for the camera motion. In summary, this experiment demonstrates that the proposed fingertip tracking method can be used for camera pose estimation without losing significant accuracy compared to a state-of-the-art fiducial marker system.