Reducing the depth data fluctuation error of the binocular imaging system based on the trapezoidal body calibration...

In binocular vision systems used in industrial settings, accurate calibration of intrinsic and extrinsic parameters is a prerequisite for ensuring the accuracy of subsequent three-dimensional positioning and dimensional measurement. However, traditional multi-image checkerboard calibration methods rely on a large number of shooting angles. The large cumulative error between images and low calibration accuracy significantly increase the data fluctuation error of the binocular ranging system, making it difficult to meet the "one-time calibration, long-term stability" requirement in industrial scenarios. To address the large error fluctuation in the ranging of small targets at medium and long distances, this study proposes a calibration method based on a single trapezoidal calibration plate. By consolidating the coordinates of all corner points in a unified coordinate system, camera parameters can be measured in a single shot, avoiding calibration failures caused by target occlusion and other issues, and significantly reducing reprojection error and depth data fluctuation. Experimental results show that this method reduces the average reprojection error to 0.31 pixels, approximately 10% lower than the classic Zhang Z. Y. method. Finally, the method is applied to a typical industrial case study: safe distance detection of 110 kV high-voltage cables. The corresponding binocular ranging system improved the standard deviation (SD) and mean square error (MSE) of depth data by 6.90% and 7.47%, respectively, at a working distance of 1.5 m, which verified its reliability and engineering practicality under industrial scenario conditions.