CN102279973B - Sea-sky-line detection method based on high gradient key points - Google Patents

Abstract

The invention discloses a sea-sky-line detection method based on high gradient key points, and the method is suitable for ocean ship object identification and positioning devices. The method is characterized in that: based on a high efficiency attention mechanism of visual information, obtaining a statistical set of high gradient key points of an image column through employing a recursion optimization algorithm and a variable resolution sampling technique, subjecting the statistical set to least squares straight line fitting, rejecting outliers which do no satisfy a distance threshold condition in the statistical set and obtaining a selected set of the high gradient key points, subjecting the selected set to least squares straight line fitting, determining a selected set of the high gradient key points which satisfies an adaptive quantity threshold and a linear correlation threshold simultaneously as sea-sky-line. A problem of effective, accurate and real-time detection of sea-sky-line in a complex sea surface environment is solved by the sea-sky-line detection method based on high gradient key points, and the method has the characteristics of strong anti-interference capability and high calculating efficiency.

Description

Sea horizon detection method based on the high gradient key point

Technical field

The invention belongs to Image processing and compute machine vision field, relate generally to a kind of naval target image-recognizing method, relate in particular to a kind of sea horizon detection method for sea target recognition and location.

Background technology

China has long shore line and wide marine territory, on the modern history of China, we " have the sea without anti-", but in today of 21st century, there is the present situation that falls behind in China naval informationization technology, still cause us " to have the perils of the sea to prevent ", can not adapt to the needs of the sacred marine territory of defendance China under the modernized information war.Under this background, research naval target image recognition technology, realize the advanced information processing technology be following high-tech war in the urgent need to.In the military field of ocean, naval target is important military target, it also is the main object of scouting and hitting, and remote naval target all is to appear near the sea horizon usually, target can be carried out potential zone location by detecting sea horizon, dwindle the scope of target search, reduce the calculated amount of target identification, therefore the sea horizon detection is prerequisite and the key of sea target recognition, and the research of sea horizon detection technique has very important military significance.

Look squarely at a distance under the state on the sea, the sea Ship target imaging generally is divided into Three regions: day dummy section, sea horizon zone, sea are regional.If target occurs at a distance, then necessarily be in the sea horizon zone.Therefore in order to determine the sea horizon zone, by detecting sea horizon, can determine the position that target occurs.For These characteristics, by determining the sea horizon zone, for the calculated amount of the work such as the target detection that reduces follow-up sea and sky background image and location, suppress noise unnecessary outside the sea horizon zone and the interference of decoy has great importance.

Sea and sky junction are the larger zones of grey scale change, thereby sea horizon is that shade of gray changes more a little bigger line.Because sea horizon is a brightness by the gradation zone of high (sky) to low (sea) generally speaking, and has certain degree of tilt usually, thus can be with its refinement, fit to straight line.

Have in recent years many scholars to carry out the research of sea horizon detection method, but traditional algorithm mainly adopt image partition method to detect.Chinese periodical " applicating technology " 2006, Vol.33, No.6, pp.96-98 has published one piece and has been entitled as the paper of " complicated sea-sky-line area detection algorithm research ", the people such as author Xie Hong disclose it and have adopted maximum variance between clusters (OTSU) to carry out the achievement in research that sea horizon detects in this paper, Fig. 1 has provided the experimental result that adopts this detection method.Fig. 1 (a) is original image, and Fig. 1 (b) is maximum variance between clusters OSTU Threshold segmentation image, and Fig. 1 (c) is at original image stack OSTU cut-off rule.As seen from Figure 1, it is unsatisfactory to adopt maximum variance between clusters to carry out the detection of sea horizon, this is because it is a kind of threshold segmentation method of integral body, and it adopts single threshold value to carry out image segmentation by the intensity profile characteristic of whole image, and this anti-noise ability that has determined it is relatively poor.Adopt the threshold value of an integral body to cut apart to entire image, beyond thought result can occur under many circumstances, especially may make part sea horizon deviation serious, serious deviation just occurred such as the right half part on Fig. 1 (c).Owing to after adopting maximum variance between clusters to carry out Threshold segmentation, be difficult to detect desirable sea horizon.Therefore, realize that effective sea horizon detects necessary research and seeks more effective, accurate and real-time algorithm.

Summary of the invention

The technical problem to be solved in the present invention is, overcomes the deficiencies in the prior art, propose a kind of can be in bad border of complicated sea sea horizon detection method effective, accurate, real-time, that antijamming capability is strong.

Sea horizon detection method of the present invention may further comprise the steps:

The first step, after receiving the sense command that ShipTargets is identified and locating device sends, gathering a frame resolution by corresponding sensor is the image of W*H;

Second step asks for that w is listed as corresponding high gradient key point p in the two field picture _iAnd composition high gradient key point statistics set P={p ₁, p ₂..., p _w, wherein: high gradient key point p _iPosition in image is (x _i, y _i), i=1,2 ..., w, and w≤W, described high gradient key point p _iAt same adjacent two the line segment L that list ₁And L ₂Between have the poor of maximum gradation value mean value, described adjacent two line segment L ₁And L ₂Respectively contain M pixel, and M is much smaller than H;

In the 3rd step, calculate described high gradient key point statistics set P={p according to following formula ₁, p ₂..., p _wThe correlation parameter of least square fitting straight line:

$\left\{\begin{array}{c}k=\frac{L_{\mathrm{xy}}}{L_{\mathrm{xx}}}=\frac{\underset{i=0}{\overset{w-1}{\mathrm{\Σ}}}(x_i-\stackrel{\‾}{x})(y_i-\stackrel{\‾}{y})}{\underset{i=0}{\overset{w-1}{\mathrm{\Σ}}}{(x_i-\stackrel{\‾}{x})}^2}\\ b=\stackrel{\‾}{y}-a\stackrel{\‾}{x}\end{array}\right.$

Wherein:

I=1,2 ..., w

In the formula, k, b are respectively the slopes of fitting a straight line y (x)=kx+b and cut square;

In the 4th step, calculate each high gradient key point P in the set of described high gradient key point statistics according to following formula _iFrom the fitting a straight line y of institute (x)=kx+b apart from dist:

$\mathrm{dist}=\frac{\mathrm{kx}-y+b}{\sqrt{k^2+1}}$

The 5th step is with each the high gradient key point p that obtains _iApart from comparing apart from dist and distance threshold D of fitting a straight line, with the high gradient key point p that satisfies dist＜D _iForm the selected set of high gradient key point P={p ₁, p ₂..., p _n, p wherein _i=(x _i, y _i), i=1,2 ..., n, and n≤w;

In the 6th step, judge whether the quantity n of the high gradient key point in the selected set of described high gradient key point satisfies self-adaptation amount threshold N, i.e. n＞N, if yes, carried out for the 7th step, if NO, then export original image and forwarded for the tenth step to ShipTargets identification and locating device;

In the 7th step, calculate the selected set of described high gradient key point P={p according to following formula ₁, p ₂..., p _nThe correlation parameter of least square fitting straight line:

$\left\{\begin{array}{c}{k}_1=\frac{L_{\mathrm{xy}}}{L_{\mathrm{xx}}}=\frac{\underset{i=0}{\overset{n-1}{\mathrm{\Σ}}}(x_i-\stackrel{\‾}{x})(y_i-\stackrel{\‾}{y})}{\underset{i=0}{\overset{n-1}{\mathrm{\Σ}}}{(x_i-\stackrel{\‾}{x})}^2}\\ b_1=\stackrel{\‾}{y}-a\stackrel{\‾}{x}\end{array}\right.$

Wherein: $\stackrel{\‾}{x}=\underset{i=0}{\overset{n-1}{\mathrm{\Σ}}}{x}_i/n$

$\stackrel{\‾}{y}=\underset{i=0}{\overset{n-1}{\mathrm{\Σ}}}{y}_i/n$

In the formula, k ₁, b ₁Respectively fitting a straight line y (x)=k ₁X+b ₁Slope and cut a square;

In the 8th step, calculate the selected set of described high gradient key point P={p according to following formula ₁, p ₂..., p _nThe linearly dependent coefficient r of least square fitting straight line:

$r=\frac{{L:}_{\mathrm{xy}}}{\sqrt{L_{\mathrm{xx}}{L}_{\mathrm{yy}}}}$

Wherein, $L_{\mathrm{xx}}=\underset{i=0}{\overset{n-1}{\mathrm{\Σ}}}{(x_i-\stackrel{\‾}{x})}^2$

$L_{\mathrm{yy}}=\underset{i=0}{\overset{n-1}{\mathrm{\Σ}}}{(y_i-\stackrel{\_}{y})}^2$

$L_{\mathrm{xy}}=\underset{i=0}{\overset{n-1}{\mathrm{\Σ}}}(x_i-\stackrel{\‾}{x})(y_i-\stackrel{\‾}{y})$

Linearly dependent coefficient r and linear dependence threshold value R are compared, if r＞R judges the selected set of high gradient key point P={p ₁, p ₂..., p _nFor the sea horizon that detects and carry out next step, otherwise to ShipTargets identification with locating device output original image and forwarded for the tenth step to;

In the 9th step, on a described two field picture, mark and draw sea horizon with straight line, and export this two field picture to ShipTargets identification with locating device;

The tenth step, judge to have or not shutdown command, if NO, then repeat nine steps of the first step to the, if yes, detection of end.

According to the present invention, ask for described high gradient key point p _iConcrete steps be: two adjacent segments L that calculate H pixel in the i row ₁And L ₂Gray-scale value mean value poor, calculate first and adopt following formula:

$d_1=\frac{\underset{j=1}{\overset{M}{\mathrm{\Σ}}}f\left(L_2\left(j\right)\right)-\underset{j=1}{\overset{M}{\mathrm{\Σ}}}f\left(L_1\left(j\right)\right)}{M}$

In the formula, f (L ₁(j)), f (L ₂(j)) be respectively the 1st adjacent two the line segment L of pixel ₁And L ₂On pixel L ₁(j) and L ₂(j) corresponding gray-scale value; d ₁Be the 1st adjacent two the line segment L of pixel ₁And L ₂Gray-scale value mean value poor; Calculate for the second time and later on and adopt following formula:

$d_j=d_{j-1}+\frac{f_{\mathrm{new}}-f_{\mathrm{old}}}{M},$ j＝2，3，…，H

In the formula, f _NewBe j adjacent two the line segment L of pixel ₁And L ₂The gray-scale value of newly-increased pixel after pixel of picture altitude direction translation, f _OldBe j adjacent two the line segment L of pixel ₁And L ₂Reject the gray-scale value of pixel after pixel of picture altitude direction translation; d _jBe j adjacent two the line segment L of pixel ₁And L ₂Between gray-scale value mean value poor;

After the whole calculating of H pixel are complete, adopt following formula to find the solution:

d _max＝max(d _j)，j＝1，2，…，H

d _MaxPoor for this row maximum gradation value mean value, its corresponding pixel position p _Max(x _Max, y _Max) be the high gradient key point position of these row in the image.

According to the present invention, as described high gradient key point statistics set P={p ₁, p ₂..., p _wIn high gradient key point p _iBe evenly distributed on the width of a described two field picture, w=W/m and m are the pixel number at minute intervals such as picture traverse W.

According to the present invention, getting described distance threshold D is 5% of a described two field picture height H; Described self-adaptation amount threshold N is 30%～60% of w; Described linear dependence threshold value R is between 0.8～0.9.

Overall technology effect of the present invention is embodied in the following aspects.

(1) the present invention is based on the efficient attention mechanism of visual information, at first seek the statistics set of the high gradient key point in the two field picture row, and the least square line match is carried out in set to high gradient key point statistics.Do not satisfy after the point not in the know of distance threshold condition by rejecting, to obtain the selected set of high gradient key point.After selected set re-starts the least square line match to the high gradient key point again, the selected set of high gradient key point of satisfying simultaneously self-adaptation amount threshold and linear dependence threshold value is judged to be sea horizon, in this two field picture, the sea horizon that detects is marked and drawed and exported to ShipTargets with image and identify and locating device.Because detection method of the present invention meets the natural current conditions in day environment of sea, thereby, comparing with other existing sea horizon detection method, the present invention has precision height, characteristics that real-time is good.

(2) in the present invention, the statistics set of high gradient key point is to adopt the recursion optimized algorithm, its Basic practice is, each of high gradient key point lists in asking for image, pixel gray-scale value to two adjacent line segments of each point is not to carry out whole adding up, but contrast grey value difference on adjacent two line segments according to the principle of from the top down translation, remove the gray-scale value of pixel after adding the gray-scale value of newly-increased pixel and deducting translation.This recursion optimized algorithm has been avoided the needed huge calculated amount of method of exhaustion, has reduced to a great extent calculation cost of the present invention.

(3) in the present invention, the quantity of high gradient key point is based on the variable resolution Sampling techniques and determines in the set of the statistics of high gradient key point, namely the horizontal direction decrease resolution in the two field picture, not only save computing time, and avoided around the sea horizon sea-surface target to the interference of detection algorithm.Like this, even image is very large, do not have much changes through sea horizon detection algorithm after the variable resolution processing in the requirement of calculated amount, and the uniform sampling method of this variable resolution can't the effect characteristics statistical accuracy, improved to a great extent on the contrary counting yield, and this is vital for real-time computer vision system.

(4) in the present invention, the self-adaptation amount threshold is set to the number percent of contained high gradient key point quantity in the set of high gradient key point statistics, this self-adaptation amount threshold that will judge the selected set of high gradient key point is arranged to the method that adapts with high gradient key point statistical value, greatly increase adaptive ability and the flexibility ratio of sea horizon detection algorithm of the present invention, reduced the dependence of sea horizon detection algorithm to image size and high gradient key point statistics number.

Description of drawings

Fig. 1 is the sea horizon test experience result of existing maximum variance between clusters image segmentation.

Fig. 2 is the operational flowchart of sea horizon detection method of the present invention.

The high gradient key point of optimizing based on recursion among Fig. 3 the present invention is extracted synoptic diagram.

Fig. 4 selects synoptic diagram based on the high gradient key point of variable resolution sampling among the present invention.

Fig. 5 is based on the least square method sea horizon test experience result of a plurality of high gradient key points.

Embodiment

The present invention is described in further detail below in conjunction with accompanying drawing and preferred embodiment.

Sea horizon shows as a statistical significance epigraph gray-scale value line jumpy in image, point on the sea horizon is normally perpendicular to changing maximum point on the gradation of image Data-Statistics meaning in the row pixel of sea horizon, these points are exactly the high gradient pixel in the image, are referred to as in the present invention the high gradient key point.In a lot of situations, when target is positioned on the sea horizon or near the sea horizon time, relevant position shade of gray in the horizontal direction will significantly become greatly, also to some extent variation of shade of gray shows as the upper and lower edge of target, and no longer is sea horizon on the vertical direction.These variations cause the rectilinearity of sea horizon destroyed.Equally, the existence of large tracts of land cloud cluster detects impact obviously to sea horizon, because cloud cluster also is the violent place of grey scale change, easily makes detected sea horizon deflection sky.In addition, complicated sea clutter is also obscured the edge of wave of the detection, particularly high gradient of sea horizon to some extent, can make the sea horizon dislocation that detects.

The sea horizon detection method that the preferred embodiment of the present invention provides detects the software package realization by storer, image pick-up card and sea horizon are housed.Distance threshold D, self-adaptation amount threshold N and linear dependence threshold value R have been deposited in the storer.The function that sea horizon detects software package is, finishes the real-time detection of sea horizon according to workflow shown in Figure 2, and its testing process comprises following three parts content.

One, obtains the statistics set of high gradient key point

On computers behind the electricity, after receiving the sense command that ShipTargets is identified and locating device provides, sea horizon detects software package, and at first to obtain the frame resolution that respective sensor collects by image pick-up card be the image of W*H, and adopt the recursion optimized algorithm and calculate high gradient key point statistics set P={p in this two field picture based on the variable resolution Sampling techniques ₁, p ₂..., p _wAnd i=1,2 ..., w, high gradient key point p _iPosition in image is (x _i, y _i).

Because the impact of optical imagery characteristic and noise in image, the gray-value variation between adjacent two pixels exists very large randomness.In order to improve these high gradient key points p _iThe accuracy and the reliability that detect, the present invention lists two adjacent segments L by calculating certain pixel same ₁And L ₂Gray-scale value mean value determine high gradient key point (referring to Fig. 3) in these row.Find the solution high gradient key point p _iThe specific implementation step be, list at the i of a two field picture, centered by first pixel, along the picture altitude direction two adjacent line segment L are set ₁And L ₂, they all respectively comprise M pixel.Wherein, M is taken as between 5～12, if the value of M is too small, what then may detect is noise spot; If the value of M is excessive, then may can't detect the high gradient key point.In this preferred embodiment, get M=8.Calculate first adjacent two line segment L ₁And L ₂Gray-scale value mean value adopt following formula:

$d_1=\frac{\underset{j=1}{\overset{M}{\mathrm{\Σ}}}f\left(L_2\left(j\right)\right)-\underset{j=1}{\overset{M}{\mathrm{\Σ}}}f\left(L_1\left(j\right)\right)}{M}---\left(1\right)$

(1) in the formula, f (L ₁(j)), f (L ₂(j)) be respectively the 1st adjacent two the line segment L of pixel ₁And L ₂On pixel L ₁(j) and L ₂(j) corresponding gray-scale value; d ₁Be the 1st adjacent two the line segment L of pixel ₁And L ₂Gray-scale value mean value poor.

Because the height of image is H, then each pixel needs to carry out the calculating of above-mentioned formula in H pixel of the row of one in the image, and calculating each time needs to make 2 * M sub-addition, 1 subtraction and 1 division, and wherein additional calculation occupies main computing time.Then each additional calculation that lists H point is 2 * M * H time.In order to save computing time, because one lists adjacent two line segment L of certain pixel ₁And L ₂It is repetition that N-1 point arranged in the calculating of (1) formula of employing, and only having a pixel is the calculation level that need to again add, and therefore can adopt the recursion optimization to calculate, thereby avoid the huge calculated amount of method of exhaustion.The basic ideas of recursion optimized algorithm are, when asking for the high gradient key point that each lists, to the adjacent two line segment L of each pixel ₁And L ₂On the pixel gray-scale value be not carry out whole cumulative, but according to two adjacent segments from the top down the principle of a pixel of translation contrast adjacent two line segment L ₁And L ₂On difference, then for the second time and after calculate adjacent two line segment L ₁And L ₂Gray-scale value mean value adopt following formula:

$d_j=d_{j-1}+\frac{f_{\mathrm{new}}-f_{\mathrm{old}}}{M},$ j＝2，3，…，H (2)

(2) in the formula, f _NewBe adjacent two line segment L ₁And L ₂The gray-scale value of newly-increased pixel after pixel of picture altitude direction translation, f _OldBe adjacent two line segment L ₁And L ₂Reject the gray-scale value of pixel after pixel of picture altitude direction translation; d _jBe j adjacent two the line segment L of pixel ₁And L ₂Between gray-scale value mean value poor.

When one list H pixel all calculate complete after, then find the solution the difference d that this lists maximum gray-scale value mean value _Max:

d _max＝max(d _j)，j＝1，2，…，H (3)

The difference d of the maximum average value of seeking out by (3) formula _MaxCorresponding pixel position p _Max(x _Max, y _Max) be the high gradient key point position of the row of j in the image.

In a two field picture, its each row can adopt the recursion optimization to obtain a high gradient key point, to be used for carrying out the fitting a straight line of sea horizon.That is to say for the two field picture that picture traverse is W, just have W row need to carry out above-mentioned recursion optimization and calculate, its calculated amount can be very large, and simultaneously, it is also very large also can to cause the following adopted least square method to carry out the calculated amount of fitting a straight line.For this reason, the present invention adopts the sampling method selection portion apportion based on variable resolution to carry out the calculating of high gradient key point, to reduce calculated amount, as shown in Figure 4.Its concrete performing step is the width W of a two field picture to be carried out five equilibrium with an interval m pixel obtain w point, the i.e. intersection point of horizontal direction solid line and vertical direction solid line among Fig. 3; So, w is exactly the quantity that needs to carry out the high gradient key point of characteristic statistics among the present invention, and in other words, the present invention just finds the solution the high gradient key point to the row employing recursion optimized algorithm at w some place.Process through this resolution decreasing, even image is very large, reducing after m times, the calculated amount that its sea horizon detects is compared with the calculated amount that the general pattern sea horizon detects does not have much changes, and the uniform sampling method of this variable resolution can't the effect characteristics statistical accuracy, improved to a great extent on the contrary counting yield, and this point is vital for real-time computer vision system.In this preferred embodiment, the width W of image=600 (pixel), the height H of image=440 (pixel); Get m=20, obtain w=W/m=30.

Two, obtain the selected set of high gradient key point

Statistics set P={p in the high gradient key point of above-mentioned acquisition ₁, p ₂..., p _wIn, some high gradient key point is by the formation of noise, these points can affect the accuracy of detection of sea horizon, therefore, need to be rejected these points not in the know by the linear fit algorithm.

To w high gradient key point p _i(x _i, y _i) linear fit be actually the linear regression problem, can come these p of match with straight line y (x)=kx+b _j(x _i, y _i).

For this reason, calculate first two parameters of this fitting a straight line by following formula:

$\left\{\begin{array}{c}k=\frac{L_{\mathrm{xy}}}{L_{\mathrm{xx}}}=\frac{\underset{i=0}{\overset{w-1}{\mathrm{\Σ}}}(x_i-\stackrel{\‾}{x})(y_i-\stackrel{\‾}{y})}{\underset{i=0}{\overset{w-1}{\mathrm{\Σ}}}{(x_i-\stackrel{\‾}{x})}^2}\\ b=\stackrel{\‾}{y}-a\stackrel{\‾}{x}\end{array}\right.---\left(4\right)$

Wherein:

I=1,2 ..., w

(4) in the formula, k, b are respectively the slopes of institute's fitting a straight line and cut square.

Then, calculate each high gradient key point p by following formula _i(x _i, y _i) from the fitting a straight line y of institute (x)=kx+b apart from dist:

$\mathrm{dist}=\frac{\mathrm{kx}-y+b}{\sqrt{k^2+1}}---\left(5\right)$

With each high gradient key point p _iCoordinate position (x _i, y _i) replace (x, y) in (5) formula just can obtain the one group distance value corresponding with each high gradient key point.

If some high gradient key point from fitting a straight line apart from dist greater than distance threshold D, then these points are to belong to noise, should be with it rejecting, because can affect more like this precision of the straight line parameter of calculating away from the point of straight line.If some high gradient key point from straight line apart from dist less than or equal to distance threshold D, then keep these high gradient key points, and consist of a new set P={p who contains n high gradient key point ₁, p ₂..., p _n, p wherein _i=(x _i, y _i), i=1,2 ..., n.This new set is called the selected set of high gradient key point in the present invention.Usually, distance threshold D is taken as 5% of picture altitude, and in this preferred embodiment, distance threshold D gets 22.

Whether the selected set of three, judging the high gradient key point is sea horizon

Next, following analyzing and processing is carried out in the selected set of the high gradient key point that the present invention need to obtain previous step, could judge whether this selected set is sea horizon.

At first, judge selected set P={p ₁, p ₂..., p _nIn the number of high gradient key point whether greater than self-adaptation amount threshold N, if greater than self-adaptation amount threshold N, then carry out following respective handling, otherwise, finish the processing of this two field picture, export original image and begin collection and the processing of next frame image to ShipTargets identification and locating device.

Because the present invention adopts the statistics of carrying out the high gradient key point based on the variable resolution Sampling techniques, therefore, judge that whether the number that satisfies distance threshold condition point is inequitable greater than some fixing threshold values.Self-adaptation amount threshold N of the present invention is that the number percent of the high gradient key point number w in gathering according to high gradient key point statistics comes self-adaptation to determine.Be N be set to w 30%～60%.If the value of N is too small, then these points might be noise spots, and the sea horizon of trying to achieve can be unstable; If the value of N is excessive, then because the point that satisfies condition does not reach the N value, may can not get needed sea horizon.In this preferred embodiment, N is set to 40% of w.This self-adaptation amount threshold that will judge the selected set of high gradient key point is arranged to the method that adapts with high gradient key point statistical value, greatly increase adaptive ability and the flexibility ratio of sea horizon detection algorithm of the present invention, reduced the dependence of sea horizon detection algorithm to image size and high gradient key point statistics number.

Secondly, in the satisfied situation greater than self-adaptation amount threshold N of the selected set of high gradient key point, to the high gradient key point P={p in the selected set of high gradient key point ₁, p ₂..., p _nAgain fit to straight line y (x)=k ₁X+b ₁, namely calculate the slope k of this fitting a straight line with following formula ₁With a section square b ₁:

$\left\{\begin{array}{c}{k}_1=\frac{L_{\mathrm{xy}}}{L_{\mathrm{xx}}}=\frac{\underset{i=0}{\overset{n-1}{\mathrm{\Σ}}}(x_i-\stackrel{\‾}{x})(y_i-\stackrel{\‾}{y})}{\underset{i=0}{\overset{n-1}{\mathrm{\Σ}}}{(x_i-\stackrel{\‾}{x})}^2}\\ b_1=\stackrel{\‾}{y}-a\stackrel{\‾}{x}\end{array}\right.---\left(6\right)$

Wherein: $\stackrel{\‾}{x}=\underset{i=0}{\overset{n-1}{\mathrm{\Σ}}}{x}_i/n$

$\stackrel{\‾}{y}=\underset{i=0}{\overset{n-1}{\mathrm{\Σ}}}{y}_i/n$

Next, judge that whether the linearly dependent coefficient r of the selected set of high gradient key point is greater than linear dependence threshold value R, if judged result is yes, think that then the selected set of high gradient key point is the sea horizon that detects, at this moment, mark and draw a two field picture that demonstrates sea horizon to ShipTargets identification and locating device output with straight line, otherwise, this frame original image exported.After this carry out the acquisition and processing of next frame image.

Selected set can go out straight line with least square fitting for the high gradient key point, but some high gradient key point is away from fitting a straight line (for example noise spot or briming), and some high gradient key point is just very close to straight line, and this just needs a criterion.Related coefficient is that it can reflect the level of intimate of one group of data linear dependence, namely is defined as to the criterion of the linear dependence degree of institute's fitting a straight line:

$r=\frac{L_{\mathrm{xy}}}{\sqrt{L_{\mathrm{xx}}-{L:}_{\mathrm{yy}}}}---\left(7\right)$

Wherein, $L_{\mathrm{xx}}=\underset{i=0}{\overset{n-1}{\mathrm{\Σ}}}{(x_i-\stackrel{\‾}{x})}^2$

$L_{\mathrm{yy}}=\underset{i=1}{\overset{n-1}{\mathrm{\Σ}}}{(y_i-\stackrel{\‾}{y})}^2$

$L_{\mathrm{xy}}=\underset{i=0}{\overset{n-1}{\mathrm{\Σ}}}(x_i-\stackrel{\‾}{x})(y_i-\stackrel{\‾}{y})$

(7) in the formula, r is the linearly dependent coefficient of the selected set of high gradient key point.The r absolute value is more close to 1, and the linear relationship of the selected set of expression high gradient key point is better, and the r of linear relation is 1.Related coefficient is close to 0, and the linear relation of the selected set of expression high gradient key point is very poor, perhaps not is straight line.Therefore, after the fitting a straight line, usually to calculate related coefficient, be used for weighing the linear relation degree.If r is greater than certain linear dependence threshold value R, then the linear relationship of straight line is satisfied in the selected set of high gradient key point, and the straight line that simulates is more satisfactory, and wherein linear dependence threshold value R is set between 0.8～0.9.In this preferred embodiment, R is set to 0.85.

More than three partial contents are sea horizon testing processes that a two field picture is carried out, thereby in application process of the present invention, said process carries out repeatedly, until ShipTargets identification finishes after providing shutdown command with locating device.

Fig. 5 has provided and has adopted the present invention that image is carried out the experimental result that sea horizon detects, wherein, the resolution of Fig. 5 (a) and Fig. 5 (b) is 600*440 (pixel), and the resolution of Fig. 5 (c) and Fig. 5 (d) is 320*240 (pixel).Black line among the figure represents to carry out the resulting fitting a straight line of fitting a straight line with the statistics set of high gradient key point for the first time; White straight line among the figure represents that the selected set with the high gradient key point re-starts fitting a straight line and a new fitting a straight line obtaining.Fig. 5 (a) and for the first time match of Fig. 5 (c) expression and for the second time match are the same straight lines, and straight line overlaps, and is w=n therefore, without the situation of point not in the know.The match second time of Fig. 4 (b) and for the first time match of Fig. 4 (d) expression and eliminating point not in the know is not the same straight line, is w＜n therefore, and the situation of point not in the know is arranged.Fig. 5 (a) is the sea horizon testing result for Fig. 1 (a) original image, and Fig. 5 (b), Fig. 5 (c) and Fig. 5 (d) are the sea horizon testing results for other image.Can find out that the present invention is because employing based on the least square method sea horizon detection method of a plurality of high gradient key points, therefore has higher accuracy of detection.

Claims (5)

1. sea horizon detection method based on the high gradient key point, it is characterized in that: the method may further comprise the steps:

The first step, after receiving the sense command that ShipTargets is identified and locating device sends, gathering a frame resolution by corresponding sensor is the image of W*H;

Second step asks for that w is listed as corresponding high gradient key point p in the two field picture _iAnd composition high gradient key point statistics set P={p ₁, p ₂..., p _w, wherein: high gradient key point p _iPosition in image is (x _i, y _i), i=1,2 ..., w, and w≤W, described high gradient key point p _iAt same adjacent two the line segment L that list ₁And L ₂Between have the poor of maximum gradation value mean value, described adjacent two line segment L ₁And L ₂Respectively contain M pixel, and M is much smaller than H;

In the 3rd step, calculate described high gradient key point statistics set P={p according to following formula ₁, p ₂..., p _wThe correlation parameter of least square fitting straight line:

Wherein:

I=1,2 ..., w

In the formula, k, b are respectively the slopes of fitting a straight line y (x)=kx+b and cut square;

In the 4th step, calculate each high gradient key point p in the set of described high gradient key point statistics according to following formula _iFrom the fitting a straight line y of institute (x)=kx+b apart from dist:

The 5th step is with each the high gradient key point p that obtains _iApart from comparing apart from dist and distance threshold D of fitting a straight line, with the high gradient key point p that satisfies dist＜D _iForm the selected set of high gradient key point P={p ₁, p ₂..., p _n, p wherein _i=(x _i, y _i), i=1,2 ..., n, and n≤w;

In the 6th step, judge whether the quantity n of the high gradient key point in the selected set of described high gradient key point satisfies self-adaptation amount threshold N, i.e. n＞N, if yes, carried out for the 7th step, if NO, then export original image and forwarded for the tenth step to ShipTargets identification and locating device;

In the 7th step, calculate the selected set of described high gradient key point P={p according to following formula ₁, p ₂..., p _nThe correlation parameter of least square fitting straight line:

Wherein:

In the formula, k ₁, b ₁Respectively fitting a straight line y (x)=k ₁X+b ₁Slope and cut a square;

In the 8th step, calculate the selected set of described high gradient key point P={p according to following formula ₁, p ₂..., p _nThe linearly dependent coefficient r of least square fitting straight line:

Wherein,

Linearly dependent coefficient r and linear dependence threshold value R are compared, if r＞R judges the selected set of high gradient key point P={p ₁, p ₂..., p _nFor the sea horizon that detects and carry out next step, otherwise to ShipTargets identification with locating device output original image and forwarded for the tenth step to;

In the 9th step, on a described two field picture, mark and draw sea horizon with straight line, and export this two field picture to ShipTargets identification with locating device;

The tenth step, judge to have or not shutdown command, if NO, then repeat nine steps of the first step to the, if yes, detection of end.

2. the sea horizon detection method based on the high gradient key point according to claim 1 is characterized in that: ask for described high gradient key point p _iConcrete steps be: two adjacent segments L that calculate H pixel in the i row ₁And L ₂Gray-scale value mean value poor, calculate first and adopt following formula:

In the formula, f (L ₁(j)), f (L ₂(j)) be respectively the 1st adjacent two the line segment L of pixel ₁And L ₂On pixel L ₁(j) and L ₂(j) corresponding gray-scale value; d ₁Be the 1st adjacent two the line segment L of pixel ₁And L ₂Gray-scale value mean value poor; Calculate for the second time and later on and adopt following formula:

In the formula, f _NewBe j adjacent two the line segment L of pixel ₁And L ₂The gray-scale value of newly-increased pixel after pixel of picture altitude direction translation, f _OldBe j adjacent two the line segment L of pixel ₁And L ₂Reject the gray-scale value of pixel after pixel of picture altitude direction translation; d _jBe j adjacent two the line segment L of pixel ₁And L ₂Between gray-scale value mean value poor;

After the whole calculating of H pixel are complete, adopt following formula to find the solution:

d _max＝max(d _j)，j＝1，2，…，H

d _MaxPoor for this row maximum gradation value mean value, its corresponding pixel position p _Max(x _Max, y _Max) be the high gradient key point position of these row in the image.

3. the sea horizon detection method based on the high gradient key point according to claim 1 is characterized in that: as described high gradient key point statistics set P={p ₁, p ₂..., p _wIn high gradient key point p _iBe evenly distributed on the width of a described two field picture, w=W/m and m are the pixel number at minute intervals such as picture traverse W.

4. according to claim 1 and 2 or 3 described sea horizon detection methods based on the high gradient key point, it is characterized in that: getting described distance threshold D is 5% of a described two field picture height H; Described self-adaptation amount threshold N is 30%～60% of w; Described linear dependence threshold value R is between 0.8～0.9.

5. the sea horizon detection method based on the high gradient key point according to claim 4 is characterized in that: the width W of a described two field picture=600 pixels, height H=440 pixels; Get the m=20 pixel, namely calculate described high gradient key point p _iRow w=30; Get M=20, N=12, R=0.85.

Images