binocular_disparity_ms — Compute the disparities of a rectified stereo image pair using
multi-scanline optimization.
binocular_disparity_ms(ImageRect1, ImageRect2 : Disparity, Score : MinDisparity, MaxDisparity, SurfaceSmoothing, EdgeSmoothing, GenParamName, GenParamValue : )
binocular_disparity_ms calculates the disparity between two
rectified stereo images ImageRect1 and ImageRect2
using multi-scanline optimization. The resulting disparity image is
returned in Disparity.
In contrast to binocular_distance_ms, the results are
not transformed into distance values.
For this task, the three operators binocular_disparity,
binocular_disparity_mg, and binocular_disparity_ms
can be used. binocular_disparity returns robust results
in regions of sufficient texture but fails where there is none.
binocular_disparity_mg interpolates low-texture regions
but blurs discontinuities. binocular_disparity_ms preserves
discontinuities and interpolates partially.
binocular_disparity_ms requires a reference image
ImageRect1 and a search image ImageRect2 which both
must be rectified, i.e., corresponding pixels must have the same
row coordinate. If this assumption is violated, the images
can be rectified by using the operators calibrate_cameras,
gen_binocular_rectification_map, and map_image.
ImageRect1 and ImageRect2 can have different widths
but must have the same height. Given a pixel in ImageRect1,
the homologous pixel in ImageRect2 is selected by searching
along the corresponding row in ImageRect2
and matching both pixels based on a similarity measure.
The disparity is the number of pixels
by which each pixel in ImageRect1 needs to be moved
to reach the homologous pixel in ImageRect2.
The search space is confined by the minimum and maximum disparity
values MinDisparity and MaxDisparity. If the
minimum and maximum disparity values are set to an empty tuple, they
are automatically set to the maximum possible range for the
given images ImageRect1.
To calculate the disparities from the similarity measure,
the intermediate results are optimized by a multi-scanline method.
The optimization increases the robustness in low-texture areas
without blurring discontinuities in the disparity image.
The optimization is controlled by the parameters
SurfaceSmoothing and EdgeSmoothing.
SurfaceSmoothing controls the smoothness within surfaces.
High values suppress disparity differences of one pixel.
EdgeSmoothing controls the occurrence and the shape of edges.
Low values allow many edges, high values lead to fewer and rounder edges.
For both parameters, reasonable values usually range between 0 and 100.
If both parameters are set to zero, no optimization is performed.
The calculation of the disparities can be controlled by generic
parameters. The following generic parameters GenParamName
and the corresponding values GenParamValue are supported:
Activates an optional post-processing step to increase robustness.
Concurrent direct and reverse matches between
reference patterns in ImageRect1 and ImageRect2 are
required for a disparity value to be returned.
The check is switched off by setting GenParamValue
to 'false'.
List of values: 'true', 'false'.
Default: 'true'.
Adapts the quality of the coarse-to-fine approach at discontinuities.
The higher the value set in GenParamValue, the more runtime is
required.
Suggested values: 2, 3, 4.
Default: 3.
Determines the method used to calculate the disparities. The following
parameters GenParamValue can be set:
'accurate': Most accurate calculation method, but requires more runtime and memory compared to the remaining methods.
'fast':
Uses a coarse-to-fine scheme to improve the runtime.
The coarse-to-fine scheme works in a similar way to the scheme explained
in binocular_disparity.
The coarse-to-fine method requires
significantly less memory and is significantly faster
than the 'accurate' method, especially for large
images or a large range of MinDisparity and
MaxDisparity.
The coarse-to-fine scheme has the further advantage that it
automatically estimates the range of MinDisparity and
MaxDisparity while traversing through the pyramid. As a
consequence, neither MinDisparity
nor MaxDisparity needs to be set. However, the generated
disparity images are less accurate for the 'fast' method
than for the default 'accurate' approach. Especially
at sharp disparity jumps the 'fast' method preserves
discontinuities less accurately.
'very_fast': Also uses a coarse-to-fine scheme to improve the runtime even further. However, this approach makes numerous assumptions that may lead to a smoothing of the disparities at discontinuities. Per default, the number of levels of the coarse-to-fine scheme is estimated automatically. It is possible to set the number of levels explicitly (see 'num_levels').
The runtime of the operator is approximately linear to the image width, the image height, and the disparity range. Consequently, the disparity range should be chosen as narrow as possible for large images. The runtime of the coarse-to-fine scheme (which is used for 'fast' or 'very_fast') is approximately linear to the image width and the image height. For small images and small disparity ranges the runtime of the coarse-to-fine scheme may be larger than that of the 'accurate' scheme.
List of values: 'accurate', 'fast', 'very_fast'.
Default: 'accurate'.
Determines the number of pyramid levels that are used for the coarse-to-fine
scheme. By setting GenParamValue to 'auto', the number of
pyramid levels is automatically calculated.
Suggested values: 2, 3, 'auto'.
Default: 'auto'.
Sets the similarity measure to be used. For both options 'census_dense' (default) and 'census_sparse', the similarity measure is based on the Census transform. A Census transformed image contains for every pixel information about the intensity topology within a support window around it.
'census_dense': Uses a dense 9 x 7 pixels window and is more suitable for fine structures.
'census_sparse': Uses a sparse 15 x 15 pixels window where only a subset of the pixels is evaluated. Is more robust in low-texture areas.
List of values: 'census_dense', 'census_sparse'.
Default: 'census_dense'.
Activates sub-pixel refinement of disparities when set to 'true'. Can be deactivated by setting 'false'.
List of values: 'true', 'false'.
Default: 'true'.
The resulting disparity is returned in the single-channel image
Disparity. A quality measure for each disparity value
is returned in Score, containing the best (lowest) result
of the optimized similarity measure of a reference pixel.
ImageRect1 (input_object) singlechannelimage → object (byte*) *allowed for compute devices
Rectified image of camera 1.
ImageRect2 (input_object) singlechannelimage → object (byte*) *allowed for compute devices
Rectified image of camera 2.
Disparity (output_object) singlechannelimage → object (real)
Disparity map.
Score (output_object) singlechannelimage → object (real)
Score of the calculated disparity.
MinDisparity (input_control) integer → (integer)
Minimum of the expected disparities.
Default: -30
Value range:
-32768
≤
MinDisparity
≤
32768
Restriction:
MinDisparity <= MaxDisparity
MaxDisparity (input_control) integer → (integer)
Maximum of the expected disparities.
Default: 30
Value range:
-32768
≤
MaxDisparity
≤
32768
Restriction:
MinDisparity <= MaxDisparity
SurfaceSmoothing (input_control) integer → (integer)
Smoothing of surfaces.
Default: 50
Suggested values: 20, 50, 100
Restriction:
SurfaceSmoothing >= 0
EdgeSmoothing (input_control) integer → (integer)
Smoothing of edges.
Default: 50
Suggested values: 20, 50, 100
Restriction:
EdgeSmoothing >= 0
GenParamName (input_control) attribute.name(-array) → (string)
Parameter name(s) for the multi-scanline algorithm.
Default: []
List of values: 'consistency_check', 'disparity_offset', 'method', 'num_levels', 'similarity_measure', 'sub_disparity'
GenParamValue (input_control) attribute.value(-array) → (string)
Parameter value(s) for the multi-scanline algorithm.
Default: []
Suggested values: 'accurate', 'fast', 'very_fast', 'census_dense', 'census_sparse', 'true', 'false', 'auto'
read_image (BaseballL, 'stereo/epipolar/baseball_l')
read_image (BaseballR, 'stereo/epipolar/baseball_r')
binocular_disparity_ms (BaseballL, BaseballR, Disparity, Score, \
-40, -10, 50, 50, [], [])
If the parameter values are correct, binocular_disparity_ms
returns the value 2 (
H_MSG_TRUE)
. If the input is empty (no input images are
available) the behavior can be set via
set_system('no_object_result',<Result>). If necessary, an
exception is raised.
threshold,
disparity_to_distance,
disparity_image_to_xyz
binocular_disparity,
binocular_disparity_mg,
binocular_distance,
binocular_distance_mg,
binocular_distance_ms
map_image,
gen_binocular_rectification_map,
binocular_calibration
3D Metrology