1 Hough Line Transform {#tutorial_hough_lines}
7 In this tutorial you will learn how to:
9 - Use the OpenCV functions @ref cv::HoughLines and @ref cv::HoughLinesP to detect lines in an
15 @note The explanation below belongs to the book **Learning OpenCV** by Bradski and Kaehler.
20 -# The Hough Line Transform is a transform used to detect straight lines.
21 -# To apply the Transform, first an edge detection pre-processing is desirable.
25 -# As you know, a line in the image space can be expressed with two variables. For example:
27 -# In the **Cartesian coordinate system:** Parameters: \f$(m,b)\f$.
28 -# In the **Polar coordinate system:** Parameters: \f$(r,\theta)\f$
30 ![](images/Hough_Lines_Tutorial_Theory_0.jpg)
32 For Hough Transforms, we will express lines in the *Polar system*. Hence, a line equation can be
35 \f[y = \left ( -\dfrac{\cos \theta}{\sin \theta} \right ) x + \left ( \dfrac{r}{\sin \theta} \right )\f]
37 Arranging the terms: \f$r = x \cos \theta + y \sin \theta\f$
39 -# In general for each point \f$(x_{0}, y_{0})\f$, we can define the family of lines that goes through
42 \f[r_{\theta} = x_{0} \cdot \cos \theta + y_{0} \cdot \sin \theta\f]
44 Meaning that each pair \f$(r_{\theta},\theta)\f$ represents each line that passes by
47 -# If for a given \f$(x_{0}, y_{0})\f$ we plot the family of lines that goes through it, we get a
48 sinusoid. For instance, for \f$x_{0} = 8\f$ and \f$y_{0} = 6\f$ we get the following plot (in a plane
49 \f$\theta\f$ - \f$r\f$):
51 ![](images/Hough_Lines_Tutorial_Theory_1.jpg)
53 We consider only points such that \f$r > 0\f$ and \f$0< \theta < 2 \pi\f$.
55 -# We can do the same operation above for all the points in an image. If the curves of two
56 different points intersect in the plane \f$\theta\f$ - \f$r\f$, that means that both points belong to a
57 same line. For instance, following with the example above and drawing the plot for two more
58 points: \f$x_{1} = 4\f$, \f$y_{1} = 9\f$ and \f$x_{2} = 12\f$, \f$y_{2} = 3\f$, we get:
60 ![](images/Hough_Lines_Tutorial_Theory_2.jpg)
62 The three plots intersect in one single point \f$(0.925, 9.6)\f$, these coordinates are the
63 parameters (\f$\theta, r\f$) or the line in which \f$(x_{0}, y_{0})\f$, \f$(x_{1}, y_{1})\f$ and
64 \f$(x_{2}, y_{2})\f$ lay.
66 -# What does all the stuff above mean? It means that in general, a line can be *detected* by
67 finding the number of intersections between curves.The more curves intersecting means that the
68 line represented by that intersection have more points. In general, we can define a *threshold*
69 of the minimum number of intersections needed to *detect* a line.
70 -# This is what the Hough Line Transform does. It keeps track of the intersection between curves of
71 every point in the image. If the number of intersections is above some *threshold*, then it
72 declares it as a line with the parameters \f$(\theta, r_{\theta})\f$ of the intersection point.
74 ### Standard and Probabilistic Hough Line Transform
76 OpenCV implements two kind of Hough Line Transforms:
78 a. **The Standard Hough Transform**
80 - It consists in pretty much what we just explained in the previous section. It gives you as
81 result a vector of couples \f$(\theta, r_{\theta})\f$
82 - In OpenCV it is implemented with the function @ref cv::HoughLines
84 b. **The Probabilistic Hough Line Transform**
86 - A more efficient implementation of the Hough Line Transform. It gives as output the extremes
87 of the detected lines \f$(x_{0}, y_{0}, x_{1}, y_{1})\f$
88 - In OpenCV it is implemented with the function @ref cv::HoughLinesP
93 -# **What does this program do?**
95 - Applies either a *Standard Hough Line Transform* or a *Probabilistic Line Transform*.
96 - Display the original image and the detected line in two windows.
98 -# The sample code that we will explain can be downloaded from [here](https://github.com/opencv/opencv/tree/master/samples/cpp/houghlines.cpp). A slightly fancier version
99 (which shows both Hough standard and probabilistic with trackbars for changing the threshold
100 values) can be found [here](https://github.com/opencv/opencv/tree/master/samples/cpp/tutorial_code/ImgTrans/HoughLines_Demo.cpp).
101 @include samples/cpp/houghlines.cpp
108 Mat src = imread(filename, 0);
112 cout << "can not open " << filename << endl;
116 -# Detect the edges of the image by using a Canny detector
118 Canny(src, dst, 50, 200, 3);
120 Now we will apply the Hough Line Transform. We will explain how to use both OpenCV functions
121 available for this purpose:
123 -# **Standard Hough Line Transform**
124 -# First, you apply the Transform:
127 HoughLines(dst, lines, 1, CV_PI/180, 100, 0, 0 );
129 with the following arguments:
131 - *dst*: Output of the edge detector. It should be a grayscale image (although in fact it
133 - *lines*: A vector that will store the parameters \f$(r,\theta)\f$ of the detected lines
134 - *rho* : The resolution of the parameter \f$r\f$ in pixels. We use **1** pixel.
135 - *theta*: The resolution of the parameter \f$\theta\f$ in radians. We use **1 degree**
137 - *threshold*: The minimum number of intersections to "*detect*" a line
138 - *srn* and *stn*: Default parameters to zero. Check OpenCV reference for more info.
140 -# And then you display the result by drawing the lines.
142 for( size_t i = 0; i < lines.size(); i++ )
144 float rho = lines[i][0], theta = lines[i][1];
146 double a = cos(theta), b = sin(theta);
147 double x0 = a*rho, y0 = b*rho;
148 pt1.x = cvRound(x0 + 1000*(-b));
149 pt1.y = cvRound(y0 + 1000*(a));
150 pt2.x = cvRound(x0 - 1000*(-b));
151 pt2.y = cvRound(y0 - 1000*(a));
152 line( cdst, pt1, pt2, Scalar(0,0,255), 3, LINE_AA);
155 -# **Probabilistic Hough Line Transform**
156 -# First you apply the transform:
159 HoughLinesP(dst, lines, 1, CV_PI/180, 50, 50, 10 );
163 - *dst*: Output of the edge detector. It should be a grayscale image (although in fact it
165 - *lines*: A vector that will store the parameters
166 \f$(x_{start}, y_{start}, x_{end}, y_{end})\f$ of the detected lines
167 - *rho* : The resolution of the parameter \f$r\f$ in pixels. We use **1** pixel.
168 - *theta*: The resolution of the parameter \f$\theta\f$ in radians. We use **1 degree**
170 - *threshold*: The minimum number of intersections to "*detect*" a line
171 - *minLinLength*: The minimum number of points that can form a line. Lines with less than
172 this number of points are disregarded.
173 - *maxLineGap*: The maximum gap between two points to be considered in the same line.
175 -# And then you display the result by drawing the lines.
177 for( size_t i = 0; i < lines.size(); i++ )
180 line( cdst, Point(l[0], l[1]), Point(l[2], l[3]), Scalar(0,0,255), 3, LINE_AA);
183 -# Display the original image and the detected lines:
185 imshow("source", src);
186 imshow("detected lines", cdst);
188 -# Wait until the user exits the program
197 The results below are obtained using the slightly fancier version we mentioned in the *Code*
198 section. It still implements the same stuff as above, only adding the Trackbar for the
201 Using an input image such as:
203 ![](images/Hough_Lines_Tutorial_Original_Image.jpg)
205 We get the following result by using the Probabilistic Hough Line Transform:
207 ![](images/Hough_Lines_Tutorial_Result.jpg)
209 You may observe that the number of lines detected vary while you change the *threshold*. The
210 explanation is sort of evident: If you establish a higher threshold, fewer lines will be detected
211 (since you will need more points to declare a line detected).