M3dmapAlign

This function takes two point clouds, a model point cloud and a scene point cloud, and returns the transformation that could move the scene's relative coordinate system such that the two point clouds are expressed using the same coordinates.

The function finds points in one point cloud and pairs them up with points in the other point cloud to minimize the root mean square error (RMS error) of the distance between all the paired points. From this set of point pairs, a new set of point pairs is calculated that minimizes the RMS error even more. The iterations continue, trying to reduce the RMS error in each iteration, until at least one stop condition occurs, such as when the RMS error falls below a specified threshold or the maximum number of iterations has occurred. You can specify the stop conditions using M3dmapControl().

Once a stop condition occurs, M3dmapAlign() returns the transformation that moves the scene's relative coordinate system so that the two point clouds are expressed in the same coordinates. The resulting transformation matrix can be returned in a pairwise 3D alignment result buffer (M3dmapAllocResult() with M_ALIGNMENT_RESULT) or a 4x4 MIL array buffer (MbufAlloc2d() with M_ARRAY and 32 + M_FLOAT).

Typically, you can start the iterative process by specifying a rough estimate of the final transformation as a prealignment. You can prealign the point clouds by overlapping their centroids (center of mass). In this case, you must call M3dmapControl() with M_PREALIGNMENT_MODE set to M_CENTROID. When prealigning the point clouds using their centroids, there is only a translation and no rotation. Alternatively, you can prealign the point clouds by directly passing a transformation matrix. This transformation matrix is typically either the result of a previous call to M3dmapAlign() or the result of some custom processing stored in a 4x4 MIL array buffer. You can also prealign the point clouds by combining the alignment of the centroids, which only results in a translation, with a specified transformation matrix, from which only the rotation is used.

Unlike most other functions that modify a MIL object, you can call this function concurrently from multiple threads on the same M_POINT_CLOUD_CONTAINER result buffer (ModelResult3dmapId) without using a M_MUTEX object. This is valid as long as the ModelLabelOrIndex parameter of the concurrent calls is set to a different index, and is not set to M_ALL.