Relative Yaw Drift of a Stereo Camera Pair in Two Mounting States: A Comparison Procedure With a Pre-Declared Noise-Floor Rule

Stereo depth is sensitive to the cameras' relative yaw. A procedure not found in the reviewed literature fixes in advance how one pair's drift is compared between two mounting states. Relative yaw is taken from the image motion of a near target (Z ≈ 0.85 m), which can reflect camera rotation and translation (angle plus translation/Z). Drift is the RMS of one-hour-block changes in a free-clamped-free sequence, and the state ratio is judged only if both exceed a null-test noise floor. Rules, thresholds and core analysis code were hashed before data collection. Over about 33 hours indoors on a consumer-webcam bench, a null test on the same pair gave a method repeatability (not accuracy) of 0.32″, over six times below the 2.0″ target. The RMS was 2.70″ when free, twice the 1.35″ floor, and 1.29″, below it, when clamped. The pre-declared verdict is "not resolved", and, as the rule prescribes, only the upper end of the ratio interval, 0.71, is reported. As the 95 % percentile bootstrap endpoint from the observed blocks, with drift common to successive blocks and state partly overlapping time of day, it is a descriptive statistic for this session, not a general bound.