Back to Adaptive Vision Studio website

You are here: Start » Filter Reference » Geometry 2D Fitting » LocateMultiplePointPatterns

LocateMultiplePointPatterns


Finds the occurrences of the pattern in the input points.

Name Type Range Description
inPoints Point2DArray Input points
inPattern Point2DArray Point pattern to be found
inReferenceFrame Rectangle2D* Exact position of the model object associated with the pattern in the image
inAllowRotation Bool Flag indicating whether rotation is allowed as a part of output alignment
inMinAngle Real Start of range of possible rotations
inMaxAngle Real End of range of possible rotations
inAllowScale Bool Flag indicating whether scale is allowed as a part of output alignment
inMinScale Real 0.0 - Start of range of possible scales
inMaxScale Real 0.0 - End of range of possible scales
inTilingFactor Real 0.000001 - 1.0 Defines relative size of the square tile on the plane during initial detection
inMinInitialScore Real 0.0 - 1.0 The minimum proportion of points correctly matched during initial detection
inMaxDeviation Real 0.0 - Maximal distance between two points considered matched
inMinScore Real 0.0 - 1.0 The minimum proportion of points correctly matched
inMinDistance Real 0.0 - Minimal distance between centers of two found occurrences
inDisjointObjectsOnly Bool Flag indicating whether found occurrences can have common points
outAlignments CoordinateSystem2DArray The transforms that align the input pattern to the input points
outAlignedPatterns Point2DArrayArray The aligned input pattern points
outLocatedObjects Rectangle2DArray Bounding rectangles of the found pattern occurrences

Description

The filter finds locations of a pattern of points in the set of input points. The possible rotations and scales of the found occurrences can be fully controlled using proper values of inAllowRotation, inMinAngle, inMaxAngle, inAllowScale, inMinScale and inMaxScale parameters.

The location routine consists of three phases. Only the inTilingFactor and inMinInitialScore parameters have an effect on the initial phase. Internally, the whole plane is then divided into square tiles which size depends on inTilingFactor and the average distance between two points from inPoints. A transformation becomes a candidate to be a valid pattern occurrence if at least inMinInitialScore fraction of pattern points reside in the right tiles. The candidate transformation proceeds to the second phase, where it is refined to be possibly best fitted to the data points. The result transformation is considered to be a valid output alignment if at least inMinScore fraction of aligned pattern points are at most inMaxDeviation away from their closest data points.

The end phase's purpose is to select the output alignments according to the inMinDistance and inDisjointObjectsOnly parameters values. The ultimate output is chosen so no two aligned pattern centers are closer than inMinDistance from themselves and, if inDisjointObjectsOnly parameter is set, no two aligned patterns cover the same data point.

The most difficult part to achieve reliable results seems to be the proper setting of the inTilingFactor parameter. Because of its existence, the algorithm will work especially well when the pattern consists of not too many points and distances between them are more or less equal i.e. the ratio of the greatest distance between pattern points and the smallest distance between pattern points is small. If this is not the case and the default value for inTilingFactor does not work well, one should try to adjust the value keeping in mind that the greater values should be used when the pattern visible in the inPoints is distorted and the smaller values will work best when only small distortion is present. In case of further problems with choosing the right inTilingFactor value, one can also try lowering inMinInitialScore value.

The filter performance depends heavily on the number of the pattern points. Because of that fact, it is highly advisable for the pattern to be as small as possible. The performance can be poor even for patterns with more than 15 points. Note that the shape of the pattern also matters. The execution time for symmetrical patterns is generally bigger than for asymmetrical ones.

Errors

This filter can throw an exception to report error. Read how to deal with errors in Error Handling.

List of possible exceptions:

Error type Description
DomainError Input pattern is empty in LocateMultiplePointPatterns.

Complexity Level

This filter is available on Advanced Complexity Level.

Filter Group

This filter is member of LocatePointPatterns filter group.

See Also