Abstract: Objectives: Obtaining complete point clouds of both the straight and diverging rails is a prerequisite for measuring cross-track structural parameters of turnouts. However, line-structuredlight-based mobile inspection systems are generally unable to cover both the straight and diverging rails in a single scanning pass. In addition, under low-overlap, weak-texture, and repetitive-structure conditions, ambiguous correspondences are likely to occur between straight- and diverging-rail point clouds, thereby affecting the measurement accuracy of cross-track structural parameters. To address these issues, this paper proposes a structure-constrained registration method for straight- and diverging-rail point clouds of railway turnouts. Methods: First, the raw point cloud was processed by cloth simulation filtering and clustering to obtain the railhead point cloud. Then, with the centerline of the straight stock rail as the reference, a local registration coordinate system was established. Under structural constraints and within a sliding-window framework, key points such as the switch rail tip and frog tip were extracted, and coarse registration of the straight and curved rail point clouds was achieved based on the collinear segments of the stock rails. Subsequently, sleepers were used as segmentation units, and sleeper positions were identified from the point clouds before and after filtering and clustering to complete interval partitioning. Finally, fine registration was performed on each segmented interval under the constraint of the design radius of the curved rail. Results: In the point cloud registration task of 9 sets of independent turnouts of 3 different types, the average root mean square error of the proposed method was 29.894 mm with a standard deviation of 1.497 mm; the mean distance error of the switch rail tip and frog tip between the straight and curved rails was 0.683 mm with a standard deviation of 0.075 mm; the mean distance error between the tips of the frog rails was 1.231 mm with a standard deviation of 0.103 mm; and the mean distance error of the measurement points of the cross-track structural parameters was 3.488 mm with a standard deviation of 0.436 mm. The threshold-combination tests showed that, under the data conditions of this study, relatively low registration errors were obtained when the number of sleeper intervals was close to the number of corresponding sleepers. When the planar alignment radius threshold was set within 3.0-7.0 m, the registration error and computational time exhibited a favorable overall performance. Conclusions: Experimental results demonstrate that the proposed method achieves high-accuracy registration of rail point clouds from the straight and diverging routes of a simple turnout. Under identical input conditions, it yields lower overall registration errors, key point correspondence errors, and measurement errors for cross-route structural parameters on 9 sets of test data than the compared methods. The registration of one set of turnout point clouds takes less than 2 min, providing a reliable point-cloud basis for cross-route structural parameters measuring.