CN103020626A - Method and system for detecting elliptical target in image - Google Patents

Abstract

The invention discloses a method and a system for detecting an elliptical target in an image. The method comprises the following steps of: A. giving an image, and obtaining an edge image by using an edge detection or ridge detection algorithm; B. extracting line segments from the edge image, and generating a line segment list; C. randomly extracting L line segments from the line segment list, and carrying out ellipse fitting by using the L line segments, and combining an ellipse fitting result into a candidate parameter list when the ellipse fitting result passes validation; and D. repeating the step C till a stop condition is met, and selecting and outputting ellipse candidate elements being in line with conditions from the candidate parameter list. The method and the system have the advantages of fast operating rate, low requirement on computation memory, and high detection accuracy.

Description

Method and system for detecting elliptical target in image

Technical Field

The present invention relates to an image processing technology, and in particular, to a method for automatically detecting an elliptical target in an image, and a system using the method.

Background

Ellipse-type target detection is a key problem in computer vision, and the result is widely used as a primitive in image matching, recognition, analysis and understanding. Therefore, the ellipse detection has wide application scenes in the fields of industrial manufacturing, biomedicine and the like.

In analytic geometry, an ellipse can be composed of a series of points on a cartesian plane and satisfy the following implicit equation under non-degenerate conditions: f (a, x) a · x ax²+bxy+cy²+ dx + ey + f ═ 0, where a ═ a b c d e f]^T，x＝[x² xy y² x y 1]^TAnd x and y are respectively the abscissa and the ordinate of a point on the ellipse. The parameter a can be converted into a more common elliptical parameter c ═ x₀ y₀ α β θ]^TWherein x is₀And y₀The central point of the middle ellipse is respectively the abscissa and the ordinate, alpha and beta are respectively the lengths of the long half shaft and the short half shaft, and theta is the included angle between the long half shaft and the abscissa.

In computer vision, one of the most commonly used geometry recognition algorithms is the Hough Transform (Hough Transform) disclosed in US patent US3,069,654, obtained by p.v.c. Hough in 1962. The basic idea is to convert the possible geometric shapes in the image into the corresponding parameter space and accumulate them through a voting process, then find the peak value in the parameter space and output the corresponding parameter as the final detection result. One basic application of the hough transform is to find straight lines in binarized edge images.

On a two-dimensional plane, the equation of a straight line can be expressed as: r ═ xcos (ω) + ysin (ω).

Let p be the point on the straight line closest to the origin, r be the distance from the origin to p, and ω be the angle between the x-axis and the straight line formed by the origin and p. One basic flow of the detection of the straight line hough transform is as follows. First, a two-dimensional parameter space accumulator consisting of r and ω is established and cleared. At each edge point, the r value is calculated for each possible ω value and 1 is added to the corresponding accumulator element. And finally, searching a peak value in an accumulator, wherein r and omega values corresponding to the position of the peak value are parameters of a straight line in the image. If the accumulator contains multiple peaks, the image will contain multiple different lines. The accuracy of the final detection result also depends on the discrete accuracy of r and ω when the parameter space accumulator is established.

Because the ellipse equation corresponding to the elliptical target has 5 free parameters, if the traditional Hough transform is used, even if part of precision is sacrificed during parameter discretization, a huge 5-dimensional accumulator still needs to be established in a computer memory, so that the calculation time is long, and the application of the traditional Hough transform to the elliptical target detection is restricted by large memory requirement and poor precision in the calculation process.

In order to solve the problem of the traditional Hough Transform in detecting the multi-parameter geometric figure, x.lei et al propose a Random Hough Transform (RHT) method and are disclosed in "a new current detection method: randomised Hough Transform ", Pattern Recognition Letters, vol.11, pp.331-338, 1990. The basic idea of the stochastic hough transform mainly includes the following three aspects:

A. and randomly extracting n edge points from the image for fitting to obtain a point in a parameter space. The value n should be greater than or equal to the spatial dimension of the parameter, for example, n should be greater than or equal to 5 in the detection of an ellipse.

B. The accumulator is replaced by a candidate parameter table P. Each element in table P represents a point in parameter space, and each element also contains an accumulator score corresponding to the point in parameter space. And updating P every time a new fitting result is obtained. Firstly, the similarity between the new fitting coefficient and the elements in P is compared, and the most similar element is found out. If their similarity is less than a preset threshold, the new fit coefficient is incorporated into the matching element in P and the accumulator value for this element is incremented by 1. And if the element similar to the new fitting coefficient cannot be found in the P, adding the new fitting coefficient into the P as a new element, and setting the corresponding accumulator value to be 1.

C. A set of elements having an accumulator value greater than a predetermined threshold is found in the candidate parameter table P, and the subset of P represents the detected target set.

Because random sampling fitting is used in A, the calculation time is greatly reduced compared with the traversal of all edge points in the traditional Hough transform on the premise of reasonably defining the number of samples. The parameter list is used for replacing the traditional accumulator, so that the problem that the traditional Hough transform has large requirements on the storage space of a computer is solved. Meanwhile, the precision of the detection result can be controlled by the threshold of the similarity between the new fitting coefficient and the element in P.

When using a random hough transform for ellipse detection, typically 5 points at a time need to be randomly extracted for ellipse fitting. When complex images are processed and the image noise is large, a large number of invalid samples are introduced, invalid parameter calculation is caused, and invalid parameter allocation P elements are accumulated. Mclaughlin, Robert a. mclaughlin, in a format known as "randomised Hough transform: a random hough transform method using only 3 points for ellipse fitting is described in the technical report of improved ellipse detection with Comparison. It can be seen that the probability that randomly sampled 3 points fall on the same ellipse is much greater than the probability that randomly sampled 5 points fall on the same ellipse. Therefore, reducing the number of random samples is an effective way to address invalid samples and accumulations.

If the target detection image has certain prior knowledge, the obtained fitting parameters can be preliminarily screened by using the prior knowledge in the early detection stage before introducing the candidate parameter table P. For example, in ellipse detection, if the range in which the center of the ellipse may appear, or the range of the length of the major and minor axes of the ellipse, or the range of angles between the major and minor axes of the ellipse and the abscissa are known, a large number of invalid fitting parameters may be filtered out before accumulation is performed.

Disclosure of Invention

Aiming at the technical defects of low detection speed, high computer memory requirement and low detection precision in the ellipse detection process in the prior art, the invention provides a method and a system for detecting an elliptical target with high detection speed and high precision.

The invention is realized by adopting the following technical scheme: a method of detecting an elliptical object in an image, comprising the steps of:

A. giving an image, and obtaining an edge image by using an edge detection algorithm or a ridge detection algorithm;

B. extracting line segments from the edge image to generate a line segment list;

C. randomly extracting L line segments from the line segment list, performing ellipse fitting by using the L line segments, and merging ellipse fitting results into a candidate parameter list when the ellipse fitting results pass verification;

D. and C, repeating the step C until the stopping condition is met, and selecting and outputting the ellipse candidate elements meeting the condition from the candidate parameter table.

Wherein, the step of extracting line segments from the edge image specifically comprises:

thinning the edge image to change all the connected areas into a line segment with a pixel width;

finding out all bifurcation points and end points in the edge image;

for a closed curve without bifurcation points and end points in the edge image, randomly extracting a point from the closed curve as an end point;

and performing edge tracking by taking the end point as a starting point, stopping tracking when a bifurcation point or the end point is met, and storing a tracking line as a line segment into a line segment list.

Wherein, the step of extracting line segments from the edge image specifically comprises:

thinning the edge image to change all the connected areas into a line segment with a pixel width;

extracting all line segments in the edge image and storing the line segments in a line segment list;

each line segment in the line segment list is subjected to corner point detection, the line segment is cut at the corner point to form a plurality of sub-line segments, and each sub-line segment is stored in the line segment list.

In the step of fitting the ellipse by using the L line segments, a point set included in the L line segments is used as input of fitting the ellipse.

Wherein the step of fitting the ellipse using the L line segments specifically includes:

calculating the number of all L combinations in the traverse line segment list;

if the number of combinations is less than a predetermined threshold value Samp_thRespectively storing all combination modes of extracting L line segments from the line segment list into the sample list by one element, otherwise, randomly extracting Samp from the line segment list_thThe L line segments of the combination mode are stored in a sample list by one element respectively;

and sequentially extracting an element from the sample list, adding 1 to the iteration times, and performing ellipse fitting by taking the point set of the line segment group corresponding to the element as the input of the ellipse fitting.

Wherein the step of verifying the ellipse fitting result specifically comprises:

setting an ellipse parameter range according to prior knowledge;

and checking whether the current ellipse fitting result is within a set parameter range, and if so, verifying the ellipse fitting result.

Wherein the step of merging the ellipse fitting results into the candidate parameter list specifically comprises:

calculating the similarity between the current ellipse fitting result and each candidate ellipse element of the candidate parameter table;

if a candidate elliptical element matched with the elliptical fitting result exists in the candidate parameter table, combining the current elliptical fitting result with the matched candidate elliptical element, and accumulating the corresponding accumulator value;

and if no candidate ellipse element matched with the ellipse fitting result exists in the candidate parameter table, adding the current ellipse fitting result serving as a new candidate ellipse element into the ellipse candidate list, and resetting the corresponding accumulator.

Wherein the stop condition is one of the following:

when the number of ellipse fits is greater than or equal to a threshold;

or, when the accumulator score of one or more candidate ellipse elements in the candidate parameter table is greater than or equal to a threshold;

or, traverse all elements in the sample list.

Wherein, the step of selecting and outputting the ellipse candidate elements meeting the conditions from the candidate parameter list specifically comprises one of the following steps:

according to the candidate in the candidate parameter tableSorting the ellipse-selecting elements from large to small according to the score value of the accumulator, and sorting the top N in the candidate parameter table_thOutputting candidate elliptical elements;

or, selecting the candidate ellipse element with the highest accumulator score from the candidate parameter table and outputting the candidate ellipse element;

or, the candidate ellipse elements corresponding to the accumulator score larger than or equal to a set threshold value are picked out from the candidate parameter table and output.

In addition, the invention also discloses an image detection system using the method for detecting the elliptical target in the image.

Compared with the prior art, the invention has the following beneficial effects:

the invention can use the prior knowledge to carry out primary screening on the ellipse fitting result under the condition that the prior knowledge is available in the application. For example, in fetal head detection, it can be known that the central point of the ellipse is within the image range, the length of the radius of the major and minor axes of the ellipse does not exceed half of the length of the diagonal line of the image, and the ratio of the length of the radius of the major and minor axes can be limited within a certain range. Therefore, a large number of invalid ellipses can be eliminated, and the subsequent calculation pressure is reduced. Therefore, the method for detecting the elliptical target in the detected image has the advantages of high operation speed, small required calculation memory and high detection precision.

Drawings

FIG. 1 is a schematic flow diagram of one embodiment of the present invention;

fig. 2 is a schematic flow diagram of another embodiment of the present invention.

Detailed Description

The invention provides a method for detecting an elliptical target in an image, which is suitable for application occasions needing to detect the elliptical target in the image.

In a preferred embodiment shown in fig. 1, the method for detecting an elliptical target according to the present invention comprises the following steps:

step S11: an image I (hereinafter also simply referred to as "original image") that needs image processing (i.e., elliptical object detection) is acquired. If the noise in the original image is obvious, for example, the speckle noise in the ultrasonic B-mode image, the original image may be subjected to noise reduction processing. Let the number of iterations Iter be 0.

Step S12: and processing the original image by using an edge detection method to obtain a binary edge image.

In a preferred embodiment, a gradient-based edge detection method is used. Transverse gradient sobel operator g_xAnd the longitudinal gradient sobel operator g_yCan be respectively expressed as:

$g_x=\left[\begin{array}{ccc}-1& 0& 1\\ -2& 0& 2\\ -1& 0& 1\end{array}\right],$

$g_y=\left[\begin{array}{ccc}-1& -2& -1\\ 0& 0& 0\\ 1& 2& 1\end{array}\right].$

using g_xAnd g_yThe original image is convoluted to obtain two gradient images G in the horizontal axis and the vertical axis directions_xAnd G_y. Gradient modulus of $G=\sqrt{G_x^2+G_y^2}.$

Given a threshold value T of the gradient modulus_GThe edge image E of the original image I is defined as:

$E_{i,j}=\left\{\begin{array}{cc}1& G_{i,j}\&GreaterEqual;T_G\\ 0& G_{i,j}<T_G\end{array}\right.$

wherein i and j are positions of the pixel points in the horizontal axis direction and the vertical axis direction.

Threshold T of gradient modulus G_GThe image feature can be selected manually or generated automatically by a program according to the statistical information of the current image.

In a preferred approach, ridge detection is used instead of edge detection. The Ridge Detection method can be specifically found in T.Lindeberg, "Edge Detection and Ridge Detection with Automatic Scale Selection", J.computer Vision, vol.30, No.2, pp.117-154, 1998.

Step S13: extracting a link area from an edge detection result (namely an edge image E), and storing the result as a line segment list Elist; each element in the line segment list Elist represents a connected region in the edge map E. For the condition that a plurality of line segments are crossed, angular point (or inflection point) detection needs to be carried out on each line segment in the line segment list, truncation is carried out on the angular point, and the truncated sub-line segments are stored in the line segment list Elist.

In order to reduce the influence of noise margin and reduce the subsequent operation amount, a minimum line segment length threshold value T can be set_lminAnd will Elist all length less than T_lminThe line segment of (2) is removed. The method for extracting the line segment list from the edge image E comprises the following steps:

A. the edge image E typically needs to be refined so that each connected region becomes a line segment one pixel wide.

In a preferred embodiment, the edge image is refined by a morphological refinement algorithm, as described in L.Lam, et al, "methods of detail-A Comprehensive Survey", IEEE Transactions on Pattern analysis and Machine analysis, vol.14, No.9, pp.869-885, 1992.

B. All bifurcation points and end points are found in the edge image E and stored in a list EPlist.

C. Scanning each connected region in the edge image E to see whether the connected region contains a point in the list EPlist; if any point in the EPlist is not contained in a certain communication area, a point is randomly acquired from the communication area and added into the EPlist. Therefore, the problem that the closed curve has no bifurcation point and end point is solved, and the closed curve in the E can be correctly added into the line segment list.

D. Extracting an end point in the EPlist, tracking the connected region on the edge image E by taking the end point as a starting point, stopping tracking if another point in the list EPlist is encountered in the tracking process, namely a bifurcation point or an end point is encountered, and storing a point set in the tracking stroke as a line segment into a line segment list Elist. This step is repeated until all points in the EPlist are traversed.

E. And removing the burr line segment. If a certain bifurcation point in the edge image E comprises three line segments in Elist, and the length of the minimum line segment is less than a preset threshold value T_spurThe smallest line segment is removed and the remaining two line segments are merged into the line segment list Elist.

F. Remove all lengths in the line segment list ElistLess than T_lminThe line segment of (2).

In the ellipse fitting step, ellipse fitting is performed by using the line segments in the line segment list as basic units, so that each line segment in the line segment list is guaranteed to only contain edge information of one object in the image.

In a preferred scheme, each line segment in the line segment list is subjected to corner point detection, and the line segment is cut off at the corner point to be used as two line segments which are stored in the line segment list; meanwhile, a maximum line segment length threshold value T is preset_lmaxIf a segment length in the segment list is greater than T_lmaxCutting the line segment at the position with the maximum corner value, adding the line segment into the line segment list, and repeating the steps until all the line segment lengths in the line segment list are smaller than T_lmax。

In another preferred embodiment, the corner detection step can be referred to as C.Teh and R.T.Chin, "On the Detectionof Dominant Points On Digital currencies," IEEE Transactions On Pattern Analysis and machinery Analysis, vol.11, No.8, pp.859-872, 1989.

Step S14: and randomly extracting L line segments from the line segment list Elist, and adding 1 to the iteration number Iter.

In a preferred embodiment, the number of line segments L is a constant equal to or greater than 2. And acquiring a point set Plist contained in the selected L line segments as input of ellipse fitting.

Step S15: ellipse fitting was performed using the point set Plist. Specific Fitting methods can be found in A.Fitzgibbon et al, "Direct Least Square Fitting of Ellipses", IEEE Transactions on Pattern Analysis and machine understanding, vol.21, No.5, pp.476-480, 1999.

In a preferred embodiment, the output of the ellipse fitting is the ellipse parameter c ═ x₀ y₀ αβθ]^TWherein x is₀And y₀Respectively central point abscissa and ordinate of the ellipse, and alpha and betaThe length of the major semi-axis and the minor semi-axis is distinguished, and theta is the included angle between the ellipse and the abscissa.

Step S16: and preliminarily screening the current ellipse fitting result by adopting priori knowledge, and judging whether the current ellipse fitting result can pass verification.

In a preferred embodiment, the central point of the ellipse, the length of the radius of the major and minor axes, the ratio of the lengths of the radii of the major and minor axes, the angle between the major axis of the ellipse and the abscissa, and the like can be limited within a certain range. Therefore, a large number of invalid ellipses can be eliminated, and the subsequent calculation pressure is reduced.

Step S17: if the ellipse fitting result passes the preliminary verification of step S16, it is incorporated into the candidate parameter table P.

In a preferred embodiment, each element in the candidate parameter table P contains a set of ellipse parameters and their corresponding accumulator scores (i.e., accumulator values).

In a preferred embodiment, the current ellipse fitting result is compared with each candidate ellipse element in the candidate parameter table P (calculating and determining the similarity between the two elements). If a matching candidate ellipse element is found in the candidate parameter table P, combining the current ellipse fitting result and the matching element, and accumulating the corresponding accumulator value. If no matching element is found in the candidate parameter table P, the current fitting result is added as a new candidate ellipse element in the candidate parameter table P and the accumulator score is set to 1.

In a preferred embodiment, the similarity between two ellipses is determined by the difference between their parameters, as described in d.prasad and m.loung, "Clustering of ellipsoids based on the same principle of the same diagnosis: an aid toellipse detection algorithms ", 20103^rd IEEE International Conference on Computer Science andInformation Technology，vol.8，pp.292-297，2010。

In a preferred embodiment, the parameters of the merged two ellipses are defined as:

$c={\left[\frac{x_0^1+x_0^2}{2}\frac{y_0^1+y_0^2}{2}\frac{{\mathrm{\&alpha;}}^1+{\mathrm{\&alpha;}}^2}{2}\frac{{\mathrm{\&beta;}}^1+{\mathrm{\&beta;}}^2}{2}\frac{{\mathrm{\&theta;}}^1+{\mathrm{\&theta;}}^2}{2}\right]}^T,$

wherein, $c^1={\left[\begin{array}{ccccc}{x}_0^1& y_0^1& {\mathrm{\&alpha;}}^1& {\mathrm{\&beta;}}^1& {\mathrm{\&theta;}}^1\end{array}\right]}^T$ and $c^2={\left[\begin{array}{ccccc}{x}_0^2& y_0^2& {\mathrm{\&alpha;}}^2& {\mathrm{\&beta;}}^2& {\mathrm{\&theta;}}^2\end{array}\right]}^T$ first and second ellipse parameters.

In a preferred embodiment, when two ellipses are combined, their corresponding accumulator scores are incremented by 1.

In a preferred embodiment, when two ellipses are combined, their corresponding accumulator scores are the original score plus the fitness (fitness) of the new fit.

In a preferred embodiment, the fitness of an ellipse fitting result is defined as F ═ Op/Circ, where Op is the total number of points in the edge image E that fall on the boundary of the ellipse and Circ is the perimeter of the ellipse.

Step S18: and judging whether the current iteration number Iter meets the stop condition, if so, executing the step S19, otherwise, jumping to the step S14.

In a preferred embodiment, a threshold value Samp is preset_thWhen the number of iterations Iter (or the number of ellipse fits) is greater than Samp_thIf the condition is not satisfied, the stop condition is satisfied.

In a preferred embodiment, the accumulator scores in the candidate parameter table P are checked, and if there is an element whose score exceeds a predetermined threshold Acc_thAnd when the stopping condition is met, the circulation is terminated, and the corresponding elements are output as a final result.

Step S19: and selecting and outputting elements meeting the conditions from the candidate parameter table P. For example, in a preferred embodiment, the candidate ellipse elements in the candidate parameter table P are sorted from large to small according to the accumulator score, and the top N is taken_thAnd the candidate ellipse elements are taken as output. When N is present_thWhen the value is 1, it is equivalent to the element of the output candidate parameter table P with the largest accumulator score. As another example, in another preferred embodiment, an accumulator score threshold Acc is preset_thFinding out all accumulator scores greater than or equal to Acc from the candidate parameter table P_thAnd outputting it。

In a preferred embodiment shown in fig. 2, the method for detecting an elliptical target according to the present invention comprises the following steps:

steps S21 to S23 are the same as steps S11 to S13, respectively, and are not repeated.

Step S23 a: a sample list Q is generated.

In a preferred embodiment, each element in the sample list Q is a one-dimensional array containing the index numbers of a set of line segments in the line segment list.

In a preferred embodiment, let L be the number of a group of segments extracted at random each time. Calculating the number n of all L combinations in the traverse line segment list_samp：

$n_{\mathrm{samp}}=\frac{{\mathrm{Num}}_e!}{({\mathrm{Num}}_e-L)!L!},$

Wherein Num_eThe number of line segments in the line segment list. If the number of combinations n_sampLess than a predetermined threshold value Samp_thAnd storing all combination modes of L line segments extracted from the line segment list Elist into a sample list Q. If the number of combinations n_sampGreater than a threshold value Samp_thRandom extraction of Samp from the line segment list Elist_thThe type combination samples (one combination sample for each combination mode of the L line segments) are stored in the sample list Q.

In a preferred embodiment, the number of segments randomly extracted at a time may be a random number.

Step S24: one element is sequentially extracted from the sample list Q and 1 is added to the number of iterations Iter. In a preferred embodiment, one element in the sample list is a one-dimensional array, which includes the index numbers of a group of line segments in the line segment list Elist.

Step S25: and acquiring a line segment group corresponding to the current element, extracting a point set of the line segment group and the point set and storing the point set in Plist as input of ellipse fitting.

Step S26 to step S27 are similar to step S15 to step S17, respectively, and are not repeated.

Step S28: and judging whether the current iteration number Iter meets the stop condition, if so, executing the step S29, otherwise, jumping to the step S24.

In a preferred embodiment, when the number of iterations Iter equals Samp_thThe stop condition is satisfied when (i.e., traversing all elements in the sample list Q), otherwise the stop condition is not satisfied.

In a preferred embodiment, the accumulator scores in the candidate parameter table P are checked, and if there is an element whose score exceeds a predetermined threshold Acc_thAnd when the stopping condition is met, the circulation is terminated, and the corresponding elements are output as a final result.

Step S29 is similar to step S19 and is not repeated.

Therefore, under the condition of prior knowledge in application occasions, the method can use the prior knowledge to carry out primary screening on the ellipse fitting result. For example, in fetal head detection, it can be known that the central point of the ellipse is within the image range, the length of the radius of the major and minor axes of the ellipse does not exceed half of the length of the diagonal line of the image, and the ratio of the length of the radius of the major and minor axes can be limited within a certain range. Therefore, a large number of invalid ellipses can be eliminated, and the subsequent calculation pressure is reduced.

The above description is only for the purpose of illustrating the preferred embodiments of the present invention and is not to be construed as limiting the invention, and any modifications, equivalents and improvements made within the spirit and principle of the present invention are intended to be included within the scope of the present invention.

Claims (10)

1. A method of detecting an elliptical object in an image, the method comprising the steps of:

A. giving an image, and obtaining an edge image by using an edge detection algorithm or a ridge detection algorithm;

B. extracting line segments from the edge image to generate a line segment list;

C. randomly extracting L line segments from the line segment list, performing ellipse fitting by using the L line segments, and merging ellipse fitting results into a candidate parameter list when the ellipse fitting results pass verification;

D. and C, repeating the step C until the stopping condition is met, and selecting and outputting the ellipse candidate elements meeting the condition from the candidate parameter table.

2. The method of claim 1, wherein the step of extracting line segments from the edge image comprises:

thinning the edge image to change all the connected areas into a line segment with a pixel width;

finding out all bifurcation points and end points in the edge image;

for a closed curve without bifurcation points and end points in the edge image, randomly extracting a point from the closed curve as an end point;

and performing edge tracking by taking the end point as a starting point, stopping tracking when a bifurcation point or the end point is met, and storing a tracking line as a line segment into a line segment list.

3. The method of claim 1, wherein the step of extracting line segments from the edge image comprises:

thinning the edge image to change all the connected areas into a line segment with a pixel width;

extracting all line segments in the edge image and storing the line segments in a line segment list;

each line segment in the line segment list is subjected to angular point detection, the line segment is cut at an angular point to form a plurality of sub-line segments, and each sub-line segment is stored in the line segment list.

4. The method of claim 1, wherein the step of fitting the ellipse with the L line segments comprises using a set of points included in the L line segments as input for the fitting of the ellipse.

5. The method of claim 1, wherein the step of fitting an ellipse using L line segments specifically comprises:

calculating the number of all L combinations in the traverse line segment list;

if the number of combinations is less than a predetermined threshold

Storing all combination modes of extracting L line segments from the line segment list into the sample list by one element respectively, otherwise, randomly extracting L line segments from the line segment list

The L line segments of the combination mode are stored in a sample list by one element respectively;

and sequentially extracting an element from the sample list, adding 1 to the iteration times, and performing ellipse fitting by taking the point set of the line segment group corresponding to the element as the input of the ellipse fitting.

6. The method of claim 1, wherein the step of verifying the ellipse fitting result comprises:

setting an ellipse parameter range according to prior knowledge;

and checking whether the current ellipse fitting result is within a set parameter range, and if so, verifying the ellipse fitting result.

7. The method of claim 1, wherein the step of merging the ellipse fitting result into the candidate parameter list comprises:

calculating the similarity between the current ellipse fitting result and each candidate ellipse element of the candidate parameter table;

if a candidate elliptical element matched with the elliptical fitting result exists in the candidate parameter table, combining the current elliptical fitting result with the matched candidate elliptical element, and accumulating the corresponding accumulator value;

and if no candidate ellipse element matched with the ellipse fitting result exists in the candidate parameter table, adding the current ellipse fitting result serving as a new candidate ellipse element into the ellipse candidate list, and resetting the corresponding accumulator.

8. The method of detecting an elliptical object in an image as in claim 1, wherein the stopping condition is one of:

when the number of ellipse fits is greater than or equal to a threshold;

or, when the accumulator score of one or more candidate ellipse elements in the candidate parameter table is greater than or equal to a threshold;

or, traverse all elements in the sample list.

9. The method as claimed in claim 1, wherein the step of selecting and outputting the qualified ellipse candidate elements from the candidate parameter table comprises one of the following steps:

sorting the candidate ellipse elements in the candidate parameter table from large to small according to the score value of the accumulator to obtain the top of the candidate parameter table

Outputting candidate elliptical elements;

or, selecting the candidate ellipse element with the highest accumulator score from the candidate parameter table and outputting the candidate ellipse element;

or, the candidate ellipse elements corresponding to the accumulator score larger than or equal to a set threshold value are picked out from the candidate parameter table and output.

10. A system for detecting an elliptical object in an image, characterized in that the system uses a method for detecting an elliptical object in an image according to any of claims 1-9.

Images