Annotation

Annotation: This paper addresses the problem of validating the calibration of extrinsic parameters in object-centric multicamera systems. A method for assessing camera positioning quality is proposed, based on AprilTag fiducial markers and the leave-one-out principle. For each camera under evaluation, the characteristic points of the calibration target are triangulated from the remaining cameras and reprojected onto the evaluated camera, with the discrepancy from reference values serving as a quantitative measure of calibration quality. Two target variants are investigated: a planar target (single marker), where reference points are determined directly from detections, and a volumetric target (rigid cube with markers on each face), where reference points are recovered by solving the Perspective-n-Point (PnP) problem using the known three-dimensional geometry of the target. Two synthetic datasets with controlled conditions were generated for experimental evaluation: one with ideal visibility conditions (40 cameras, full target visibility) and one with realistic conditions (16 cameras, partial visibility). Individual random perturbations were applied to the true extrinsic parameters to simulate calibration results of varying quality. Experimental results demonstrated that both methods exhibit high Spearman rank correlation with the true positioning errors. The volumetric target method outperforms the planar target method, the intra-scene correlation was 0.943 and 0.920 respectively under ideal conditions, and 0.743 and 0.686 under realistic conditions. The gap between methods increases approximately 2.5 times as observation conditions deteriorate, indicating more robust performance of the volumetric target method in practical settings.