I've been testing ELSED on a video sequence and discovered a couple of memory overrun errors. They were all caused by wrong or missing bound checks which I have fixed here.

Hi! Your code is great ! But when I use it on my computer(intel i5-11400H Ubuntu 18.04), the speed of ELSED is about 14ms and LSD(OpenCV) is about 19ms. It is not so fast in your paper. what is wrong with my computer? I just use your code to test. I hope to get your reply. Thanks!!!!!

Hi, the code is very good! But I have some questions. In you main.cpp, detector short segments use follow params

 // Gaussian kernel size
  int ksize = 5;
  // Sigma of the gaussian kernel
  float sigma = 1;
  // The threshold of pixel gradient magnitude.
  // Only those pixels whose gradient magnitude are larger than
  // this threshold will be taken as possible edge points.
  float gradientThreshold = 30;
  // If the pixel's gradient value is bigger than both of its neighbors by a
  // certain anchorThreshold, the pixel is marked to be an anchor.
  uint8_t anchorThreshold = 8;
  // Anchor testing can be performed at different scan intervals, i.e.,
  // every row/column, every second row/column
  unsigned int scanIntervals = 2;

  // Minimum line segment length
  int minLineLen = 15;
  // Threshold used to check if a list of edge points for a line segment
  double lineFitErrThreshold = 0.2;
  // Threshold used to check if a new pixel is part of an already fit line segment
  double pxToSegmentDistTh = 1.5;
  // Threshold used to validate the junction jump region. The first eigenvalue of the gradient
  // auto-correlation matrix should be at least junctionEigenvalsTh times bigger than the second eigenvalue
  double junctionEigenvalsTh = 10;
  // the difference between the perpendicular segment direction and the direction of the gradient
  // in the region to be validated must be less than junctionAngleTh radians
  double junctionAngleTh = 10 * (M_PI / 180.0);
  // The threshold over the validation criteria. For ELSED, it is the gradient angular error in pixels.
  double validationTh = 0.15;

  // Whether to validate or not the generated segments
  bool validate = true;
  // Whether to jump over junctions
  bool treatJunctions = true;
  // List of junction size that will be tested (in pixels)
  std::vector<int> listJunctionSizes = {5, 7, 9};

when you detect large segments. what is the params？

Thanks for your nice work! I use ELSED::detectEdges() like:

upm::ELSED lineDetector;
auto edges = lineDetector.detectEdges(colorImg);
for (const auto &e: edges) {
    if (e.size() < 2)
      continue;
    const auto &startPx = e.front(), &endPx = e.back();
    // ......
}

But, I find e.back() seems not a end point of a edge! What e actually means as a vector?

Hi iago: first of all, thank you for your OPEN SOURCE. I am tring to use it in my project.but when i test it, i meet some strange problem(detect some lines that shouldn’t be exist). and finally,i add a imwrite here:

	dstInfo->gImgWO = cv::Mat:(srcImg.size(), dstInfo->dxImg.type());
	dstInfo->gImg = cv::Mat:(srcImg.size(), dstInfo->dxImg.type());
	dstInfo->dirImg = cv::Mat:(srcImg.size(), CV_8UC1);
            cv::imwrite("beforegImg.jpg ", dstInfo->gImg);

and then i find sometimes it will creat a img that is not a "black" one, some "white" pixels on it. but when i use cv::Mat::zeros instead of cv::Mat,i test many times and did't meet that again. but i can not sure this. could you help to confirm that? Thanks.