Error while using StereoBinarySGBM to compute Disparity

[TheShadow29]

System information (version)

• OpenCV => 3.1
• Operating System / Platform => Ubuntu 16.04
• Compiler => GCC 5.1

Detailed description

I am trying to use cv::stereo::SteroBinarySGBM to compute the disparity but it keeps on throwing OpenCV Error: Assertion failed (cost.cols / (maxDisparity + 1) == leftImage.cols) in hammngDistanceBlockMatching
opencv_contrib/modules/stereo/include/opencv2/stereo/matching.hpp:413: error: (-215) cost.cols / (maxDisparity + 1) == leftImage.cols in function hammingDistanceBlockMatching
The images are in grayscale and type = CV_8UC1

[mshabunin]

I tried to run example_stereo_sample aloeL.jpg aloeR.jpg and it works fine, default algorithm is STEREO_BINARY_SGM which uses an instance of cv::stereo::StereoBinarySGBM.
Please provide more info.

[TheShadow29]

@mshabunin
I am unsure what exact details I should give. Do let me know what more would be required.

I am using OpenCV and PCL in conjunction with ROS. Upto the point of the error in the code, PCL and ROS is not really being used. Everything else kept the same
This works:

cv::Ptr<cv::StereoSGBM> sgbm;
sgbm = cv::StereoSGBM::create(0,16,3)

This doesn't:

cv::Ptr<cv::stereo::StereoBinarySGBM> sgbm;
sgbm = cv::stereo::StereoBinarySGBM::create(0,16,3);

I have included the following:

#include "opencv2/calib3d/calib3d.hpp"
#include "opencv2/imgproc.hpp"
#include "opencv2/imgcodecs.hpp"
#include "opencv2/highgui.hpp"
#include "opencv2/core/utility.hpp"
#include <opencv2/core/eigen.hpp>
#include "opencv2/stereo.hpp"
#include "opencv2/photo.hpp"
#include "opencv2/imgproc/imgproc_c.h"
#include "opencv2/imgproc/detail/distortion_model.hpp"
#include "opencv2/calib3d/calib3d_c.h"
#include "opencv2/ximgproc/disparity_filter.hpp"

My sgbm parameters settings looks like this:
For the first one, I comment setBinaryKernel. Everything else remains the same.

void init_sgbm(const cv::Mat& left,int block_size = 5, int numberOfDisparities = 64)
{
	// int block_size = 5;
	num_disp = numberOfDisparities;
	wsize = block_size;
	std::cout << "line 39 \n";
	sgbm->setPreFilterCap(63);
	std::cout << "line 41 \n";
	int sgbmWinSize = block_size > 0 ? block_size : 3;
	// int numberOfDisparities = 64;
	sgbm->setBlockSize(sgbmWinSize);

	
	int cn = left.channels();

	cv::Size img_size = left.size();
	numberOfDisparities = numberOfDisparities > 0 ? numberOfDisparities : ((img_size.width/8) + 15) & -16;
	
	sgbm->setP1(8*cn*sgbmWinSize*sgbmWinSize);
	sgbm->setP2(32*cn*sgbmWinSize*sgbmWinSize);
	sgbm->setMinDisparity(0);
	sgbm->setNumDisparities(numberOfDisparities);
	sgbm->setUniquenessRatio(10);
	sgbm->setSpeckleWindowSize(100);
	sgbm->setSpeckleRange(32);
	sgbm->setDisp12MaxDiff(1);

	sgbm -> setBinaryKernelType(1);
}

[mshabunin]

Try to start with parameters from this example.

[lablabla]

I had the same problem.

It seems that when constructing the StereoBinarySGBM the params (specifically, the _numDisparities) are passed to the constructor of the parent Matching object.

When setting with the call
sgbm->setNumDisparities, the underlying Matching object's numDisparities is not being updated.

When calling compute, the
hamDist.create(left.rows, left.cols * (params.numDisparities + 1), CV_16S); line creates the matrix using StereoBinarySGBM's params.numDisparities, and the assert in
hammingDistanceBlockMatching that fails is
CV_Assert(cost.cols / (maxDisparity + 1) == leftImage.cols); is using maxDisparity of the underlying Matching

What happens is that Matching's numDisparity is 16, set when constructed, and StereoBinarySGBM's numDisparity can be different depending on numberOfDisparities in your init_sgbm function.

If they differ, it will cause the error.

What I did was to not call the setNumDisparites but to set the final numDisparity in the constructor and it solved the problem. Though the right thing to do will obviously to fix this.