Circular marker-aided multi-view laser point cloud registration based on adaptive-weighted bundle adjustment

To address the issue of cumulative error leading to poor registration results in multi-view laser point cloud registration aided by circular markers, caused by the reconstruction error of the three-dimensional (3D) coordinates of marker centres and local view transformation matrix estimation error, an Adaptive-Weighted Bundle Adjustment (AWBA) method is proposed. Firstly, coarse registration is achieved based on Euclidean distance matching and angle constraints. Then, an adaptive weighting strategy is introduced to incorporate the accuracy information into the computation of the transformation matrix to improve its estimation accuracy and inhibit the single-view registration error by considering the accuracy difference of the circle centre 3D coordinates. Next, global marker coordinates are optimised by first removing outliers from the sets of homologous points using statistical methods followed by iteratively solving the global marker coordinates using remaining weighted homologous points to improve the accuracy of the global marker. AWBA adopts a synchronous optimisation strategy to calculate the current view transformation matrix based on the latest optimised global markers when the data of a new view is acquired, and it continuously optimises the coordinates of the global marker throughout the reconstruction process to suppress the backward cumulative error. Experimental results demonstrate that AWBA enjoys state-of-the-art performance compared with other methods, with Absolute Error (AE) <0.094 mm for the standard ball radius, model Mean Absolute Distance (MAD) <0.093 mm, and Successful Registration Rate (SRR) greater than 93.010. AWBA can enhance the registration effect of multi-view laser point clouds with a wide range of applications in industrial inspection, robotic navigation and cultural heritage preservation.