Create a 3D Point Cloud using dynamic cross-correlation and polar geometry typically from a set of dots generated by a lasergrid.

Demo Profile

• CPC - T1 - Creating a 3D Model from a lasergrid_001.zip

3D reference tools (one per camera image)

• Camera 1-4 - a 3D reference is needed per camera

Point positions

You can use an external tool or a regular grid to set up the positions for creating the cloud:

2D multiple position tool

• Tool - A tool with multiple point results. Currently, these tool types are supported:  Blob2, Blob3, Blob4, TemplateFinder2, TemplateFinder3

Regular grid

• Center X/Y - ROI center
• Size X/Y - ROI size
• Count X/Y - Number of dots within ROI in X and Y directions
• In camera - Use 2D coordinate system from the selected camera

Z search

• Nominal/Range/step - search parameters for initial height search
• Template height/width - size of template (in reference coordinates)
• Search height/width - size of rectangle for cross correlation search
• Prohibit reuse of image points - a position that has already been identified as a match will not be a candidate for later matches. Helps reduce false detections
• Min accepted correlation - early exit for mismatches
• Min acepted Std ratio - ignore featureless images

3D output image

• Image - image to store and display the generated point cloud. May be deselected for string output only (below)
• Output coordinates only - only the X,Y,Z coordinates are stored
• Output coordinates, deviation and cameras used - 5-element points are stored (not currently useful)

Advanced

Match settings

• Pixel size in resampled images - change to speed up processing
• Set template size in Z scan to - change to speed up initial height search
• Stretch image contrast - optimise image contrast. This is most effective for blob search, and has less effect for correlation search
• Percent level - contrast stretch setting
• Min distance -  contrast stretch setting
• Stretch defaults - set Percent level and Min distance to default values

Correlation search

Resampled images are correlated, and maximum correlation factor determines the match position.

• Min score for match (0-255) - cross correlation match threshold
• Ignore images with lower Std ratio than - ignore featureless images
• Template selection
  • Auto - use best image as template in cross-correlation algorithm
  • Camera 1-4 - use fixed image as template

Blob search

Blobs are searched within the resampled images, and the center of gravity of a blob is used to determine the match position.

• Use thresholds - manually set thresholds
• Auto (dark) - see description below
• Auto (bright) - see description below
• Minimum threshold - for manually set thresholds
• Maximum threshold (inclusive) - for manually set thresholds
• Position factor - see description below
• Bimodal - see description below
• Morphology (e=erode, d=dilate) - erode or dilate blobs
• Blob smooth factor - blob smoothing
• Largest/Darkest/Brightest/Single blob only - only a single blob will be used for positioning; you can here select which. If you choose Single blob only, the search will fail if more blobs are within the search area.

Auto thresholding

A threshold is set based on the resampled image's histogram.

• If Bimodal is checked, the histogram should have two distinct tops (i.e., two major gray levels). A threshold is calculated assuring maximum separation.
• If Bimodal is not checked, the histogram is assumed not bimodal. A threshold is still found to separate the image, but since tops in the histogram are not assumed, the centre of gravity of the histogram on each side of the threshold is used as a "top" instead.
• The Position factor determines the reported threshold. 0 means the optimum threshold, -1 means the darker top, +1 means the brighter top. Anything in between leads to linear interpolation between a top and the threshold.

Factor for calculating deviation

There are two deviation measures calculated for each point:

• The first (aka. Deviation or 3D Deviation) measures how well the observed points coincide in 3D space. A 3D point is found as the intersection of two or more 3D lines, one line from each camera. The lines are the "inverse projections" of the 2D points into space. When two lines do not meet exactly in space, the deviation is half of the shortest distance between them. i.e., by how much the point result deviates from the lines. When more that two cameras are used, the deviations are averaged. Deviation is given in physical coordinates and reflects 3D camera calibration
• The second (2D deviation) is calculated by mapping the found 3D point back to the camera image, and measuring the distance from this point to the incoming 2D point. A large error here could mean that the tools producing the 2D points have missed.

The weight percent lets you use either deviation measure or both:

• Weight percent deviation vs. 2D deviation -  100% means use 3D deviation only; 0% means 2D deviation only. Or you can select a combination. Generally this is sound, since the 2D and 3D measurements are in the same dimension (usually mm).

Constraints

• Max deviation for accepted point - acceptance criterion for each point (see Deviation above)
• Ignore camera points with dev. ratio over [dB] - Ignore camera points with deviation ratio over [dB] - ignore cameras where 3D deviation ratio to the best camera is over the threshold - measured in dB
• Ignore camera points with deviation over - used to remove "bad" points before estimating position in space
• Minimum number of active cameras - only succeed if at least this number of cameras is accepted. This is especially useful in combination with the Template selection (above)

Outlier filter

• Filter outliers - remove points without neighbours
• Distance - distance to neighbours
• Min neighbour count - minimum count of neighbours within distance to keep point

Output string

• Create Python output tuple - Python tuple of tuples with 5 elements per sample: X, Y, Z, deviation, cameras used. Useful for scripting

To investigate tool performance, you can select a single point index and display resampled images and Z search match plots.

Select point

• Point index - index of one of the input (or grid) points. Index starts at 1
• Point position (X,Y) - the point position (in object coordinates) is displayed. This helps locating the point in the main image

Resampling

The best match in the Z scan is displayed to the left.

For Correlation search, the used template and the search areas from the other cameras are displayed to the right. The template will be the smaller image.

For Blob search, the same size is used for all resampled images. All blobs found in the images are drawn in red, and if only one blob is found, or if you select the largest/darkest/brightest as the result, it will be drawn in green. The center of gravity for this blob is the position used to locate a 3D point.

Curves

A match curve is supplied for z scan analysis. For each step in the Z scan loop, the image correlation factors are averaged and plotted (Match).

Visualisation

Axis - individual color for all axis

Results

Points in	number of points from 2D tool
Points found	number of accepted points
Outliers removed	number of outliers removed
Percent accepted	percentage of points passing
Min deviation	best fit
Max deviation	worst fit
Mean deviation	mean fit
Std deviation	standard deviation for fit
Results	optional Python string with all results