M3dregControl

Sets the technique for performing the pairwise comparison during each iteration of the registration operation, and establishes how to compute the RMS error.

In each iteration, points in the reference point cloud are paired with the closest points in the other point cloud. The points are always paired in the same manner (independent of the setting of M_ERROR_MINIMIZATION_METRIC).

Once paired, the RMS error is calculated using the specified distance measurement or error metric.

The best technique for a given registration depends on how many large and flat surfaces the object represented by the point cloud has. Using the correct setting can help avoid registration sliding (when multiple different registration are possible because of large planar surfaces) or diverging (when the registration operation results in the point clouds moving apart). INQ

Sets a stop condition for the iterative registration operation. If the maximum number of iterations is reached, the registration operation stops.

To inquire whether this stop condition was met, call M3dregGetResult() with M_STATUS.

Note that preregistration constitutes the first iteration. INQ

Sets the number of registration elements in the pairwise 3D registration context. The number of registration elements should be equal to, or greater than, the number of point clouds to register.

If you reduce the number of registration elements, the settings of the ones that remain are kept unchanged. If you increase the number of registration elements, the new ones are initialized with the default settings. INQ

Sets how to perform the preregistration step. INQ

Sets the stop condition for the registration operation that tests the percentage change of the root-mean-square (RMS) error of successive iterations. The RMS error is calculated from the distance between all paired points. You can control how the distance is calculated using M_ERROR_MINIMIZATION_METRIC.

Specifically, where E_i is the RMS error of iteration i, the relative RMS error is calculated as:

[(E_i-1 - E_i )/E_i-1 ]*100

For instance, if the RMS error of the first iteration is 2.0 and the RMS error of the second iteration is 0.8, the relative RMS error of the second iteration is equal to [(2.0-0.8)/2.0]*100, which is 60%. If the relative RMS error is positive, this means the RMS error is decreasing, and the point clouds are becoming increasingly aligned.

If the change in the calculated RMS error between successive iterations is less than or equal to the specified value, the registration operation stops.

If this stop condition is met, the registration is typically considered correct.

Note that the specified value represents a percentage, so the default value of 0.1 represents 0.1%, not 10%. INQ

Sets the stop condition for the registration operation that tests the root-mean-square (RMS) error of successive iterations. The RMS error is calculated from the distance between all paired points. You can control how the distance is calculated using M_ERROR_MINIMIZATION_METRIC.

This value should be set to an acceptable average RMS error for a pair of points.

If the calculated RMS error goes under or equals the specified value, the registration operation stops.

If this stop condition is met, the registration is typically considered correct. INQ

Sets whether to enable subsampling before applying the registration.

Settings for subsampling techniques are defined in a subsampling context. Every registration context contains a subsampling context.

To change the settings of the subsampling context, call M3dregInquire() with M_SUBSAMPLE_CONTEXT_ID to first get the MIL identifier of the subsampling context. The subsampling context can then be controlled using M3dimControl() (see the settings available for Subsample context ID). INQ

Sets the maximum amount of time for M3dregCalculate() to complete the registration operation before generating a time-out error.

Sets the percentage of point pairings to use in a single iteration of the registration operation.

At the start of an iteration, the points of the reference point cloud are individually paired with the closest points in the other point cloud. M_OVERLAP specifies the percentage of those paired points that are used in the registration operation, prioritizing the paired points with the lowest RMS error.

For example, if you specify 80%, then during each iteration, 80% of paired points (with the lowest calculated RMS error as defined by M_ERROR_MINIMIZATION_METRIC) are used to calculate the average RMS error of the iteration, and a new transformation. This excludes the 20% worst point pairings, which often contain outliers in the reference point cloud. These outliers are typically caused by image noise, occlusion, object deformity, or misalignment.

Typically, if there are fewer points in the reference point cloud than in the point cloud that will be registered to it, the overlap should roughly equal the ratio of the points in the reference point cloud to the points in the other point cloud.

If the density of points is the same in both point clouds, this ratio could be set to the exact ratio between the two. For example, if there are 100 points in the reference point cloud, 200 points in the point cloud that will be registered to it, and they have a similar density, then the overlap can be set to 50%. This can occur because of occlusion of the object while generating the scene point cloud.

If there are more points in the reference point cloud, this value should be set to a little under 100%. In this case, it is recommended that you use as many of the points in the reference point cloud as possible while still accounting for outliers. INQ

Sets the index of the iteration from which to copy the transformation matrix, or get results.

This setting is used by the M_INTERMEDIATE_ITERATION combination value in functions like M3dregCopyResult() and M3dregGetResult() to retrieve results from intermediate iterations, and is useful when you want to debug the registration operation or understand it in greater detail. INQ

Removes all results from the pairwise 3D registration result buffer. This does not delete the pairwise 3D registration result buffer, as is the case with M3dregFree().

Stops the current execution of M3dregCalculate() from another thread of higher priority. Any results from previous registrations become invalid and are discarded, and any completed results from the current registration operation will be available.