Straight Line Detector¶

The EasyImage straight line detector finds straight line segments in a BW8 binary image. It uses the probabilistic Hough transform to vote for candidate lines, then merges nearby collinear segments. The result is a vector of ELine ^.NET ^Python objects sorted from longest to shortest.

The detector accepts a BW8 image, an optional region of interest, and an optional frame. You can also supply only a region, without an image.

Input Image¶

The input is not a grayscale image but a binary BW8 image: the white pixels (255) are the surface on which lines are searched, and the black pixels (0) are ignored. This image can be generated from:

The skeleton of an image with EasyImage::Thinning ^.NET ^Python.
The edges of an image with ECannyEdgeDetector ^.NET ^Python.
The threshold of an image with EasyImage::Threshold ^.NET ^Python.

Note

The objects of interest must be white on a black background. If they are black, invert the image first with EasyImage::Oper ^.NET ^Python and the EArithmeticLogicOperation_Invert ^Python ^.NET operation.

Tip

The detection time grows with the number of white pixels. Running the detector directly on large white blobs, such as a thresholded image, is slow. When possible, reduce the blobs to thin structures first with the Canny edge detector or with the thinning operation: the detected lines are the same, but the detection is much faster.

The input image can also be denoised to improve the line detection.

Parameters¶

The detector uses two groups of parameters.

Hough parameters control the vote accumulator:

Parameter	Description	Default
`rho`	Distance resolution of the accumulator, in pixels.	1 px
`theta`	Angle resolution of the accumulator, in degrees.	1°
`threshold`	Minimum number of Hough votes for a candidate to qualify as a line.	150

Segment merging parameters control how raw detections are joined into final line segments:

Parameter	Description	Default
`minLineLength`	Minimum length of a detected segment, in pixels. Shorter segments are discarded.	64 px
`maxLineGap`	Maximum gap between two collinear segments for them to be merged into one.	164 px
`angleThreshold`	Maximum angular difference, in degrees, for two segments to be considered part of the same line.	35°

Inputs¶

ELineDetector::Detect ^.NET ^Python accepts an image, an optional region, and an optional frame. Supply a region to restrict detection to a specific area of the image. Supply a region alone, without an image, to detect lines directly on the region’s shape.

Detected lines are always returned in the coordinate system of the frame when one is provided.

Output¶

All overloads return std::vector<ELine>. The lines are sorted by length in descending order: the longest detected line is always at index 0.

Each ELine ^.NET ^Python carries:

Start point and end point.
Length.
Angle.

API¶

The API of the straight line detector is the ELineDetector ^.NET ^Python class:

ELineDetector::Detect ^.NET ^Python: detects straight lines and returns them sorted by length.
ELineDetector::SetRadialBinSize ^.NET ^Python: sets the distance resolution of the accumulator.
ELineDetector::SetAngularBinSize ^.NET ^Python: sets the angle resolution of the accumulator.
ELineDetector::SetMinNumVotes ^.NET ^Python: sets the minimum number of Hough votes.
ELineDetector::SetMinLineLength ^.NET ^Python: sets the minimum line segment length.
ELineDetector::SetMaxLineGap ^.NET ^Python: sets the maximum merging gap.
ELineDetector::SetAngleThreshold ^.NET ^Python: sets the maximum angular difference for merging.
ELineDetector::GetSegmentsBeforeMerging ^.NET ^Python: returns the raw segments found by the last detection, before they were merged. See Debugging the Detection.
ELineDetector::Save ^.NET ^Python: saves the detector configuration to a file.
ELineDetector::Load ^.NET ^Python: loads a detector configuration from a file.

Debugging the Detection¶

The final lines are built by merging raw segments, according to maxLineGap and angleThreshold. When a result is not what you expect (a line split in two, two distinct lines fused into one, a line missing), it is hard to tell whether the problem comes from the Hough detection or from the merging step.

ELineDetector::GetSegmentBeforeMerging ^.NET ^Python returns the raw segments found by the last call to ELineDetector::Detect, before the merging step. Comparing them with the final lines shows what the merging did:

If a raw segment is missing where you expect a line, the problem is upstream of the merging: adjust the Hough parameters (threshold, rho, theta), minLineLength, or improve the input image.
If several raw segments lie on the same physical line but were not merged, increase maxLineGap or angleThreshold.
If raw segments of different physical lines were fused, decrease maxLineGap or angleThreshold.

Note

The segments are only available after a call to Detect, and are overwritten by the next call.
The segments are expressed in the same coordinate system as the lines returned by Detect: the local frame when a frame is provided.
The segments are returned in no particular order, and are not sorted by length.

In eVision Studio¶

The Line Detector tool uses this method to display the raw segments on the image. After each detection, the tool retrieves the segments with GetSegmentBeforeMerging and draws them in its display.

In the setup view of the tool, open the Display options category of the property grid:

Show lines: draws the final detected lines, in blue. The selected line is drawn in green.
Show segments before merging: draws the raw segments returned by GetSegmentsBeforeMerging. Each segment is drawn with its own random colour, so that neighbouring segments can be told apart.

To debug a detection:

Enable Show segments before merging. Disable Show lines if the blue lines hide the segments.
Check that each expected line is covered by one or more raw segments. If not, adjust the Hough parameters or the minimum line length.
Look at how the segments group into lines. Adjust Maximum gap and the angle threshold in the merging parameters until the segments of a same line are merged, and the segments of different lines are not.
Re-enable Show lines to confirm the final result.

The same option is available in the view of the tool, as set from the setup.

Basic Usage¶

Create an ELineDetector instance. The default parameters work well for clear, long lines.
Adjust parameters if needed. Lower threshold or minLineLength to detect shorter or fainter lines. Increase maxLineGap to bridge gaps in broken lines.
Call ELineDetector::Detect with a binary BW8 image, and optionally a region and a frame.
Iterate over the result vector. The line at index 0 is the longest detected line.