Answers are grounded in this paper and cite their sources.
Marker-based stereo optical tracking can be disrupted by reflective interference and ambiguous correspondences when marker projections lie near the same epipolar line. We developed a tracking method that combines contour-based marker detection and constrained ellipse fitting with contour-derived monocular depth priors, uncertainty-normalized stereo matching, and adaptive region-of-interest (ROI) processing. In the reprojection-consistency evaluation, the proposed localization method achieved mean errors of 0.0532 and 0.0552 pixels in the left and right cameras, respectively, compared with 0.0827 and 0.0843 pixels for grayscale-centroid localization. In a separate inter-marker distance experiment involving 150 stereo observations, its mean absolute error was 0.0458 mm, compared with 0.0515 mm for OpenCV fitEllipse and 0.0551 mm for the grayscale-centroid method. In 50 stereo frames with deliberately induced ghost-marker ambiguity, the proposed matching method met the predefined frame-level correctness criterion in all 50 frames, compared with 20 frames for epipolar-only matching and 37 frames for MSRC. All expected markers were detected without false positives in the 100 images acquired for the paired illumination experiment. With 1–15 markers, adaptive ROI processing maintained a mean algorithmic processing time below 20 ms per stereo frame pair; at 15 markers, the mean was 16.52 ms and the 95th percentile was 21.20 ms. These results demonstrate improved measurement and matching performance under the tested laboratory conditions.
No takes yet. Share an insight, caveat, or question.
Chen et al. (2026) studied this question.
Synapse has enriched 10 related papers. Consider them for comparative context: