CN106875341A - Distorted image correction method and its localization method - Google Patents

Abstract

The invention provides a kind of distorted image correction method that can be used for fish eye images real time correction extremely fault image localization method.The distorted image correction method includes step：(1) correction parameter is determined；(2) according to the correction parameter, correcting algorithm is determined；Wherein step (1) further includes step：(11) position and the profile of multichannel fault image are determined；(12) correction factor of the fault image is determined.The distorted image correction method is accurately positioned using bow and beam bearings to the position of the fault image and profile, to ensure the accurate and effective of the distorted image correction method.

Description

Distorted image correction method and its localization method

Technical field

The present invention relates to the image rectification technology in photography and vedio recording field, in more detail it is related to a distorted image correction Method and its localization method.

Background technology

Photography, shooting occupy very important status in the daily life and work of modern, it has also become people An indispensable part in life and work, life.

People have been accustomed to using the dribs and drabs having in the electronic equipment record life of photography, camera function. People like and need a kind of such instrument to record the growth of child, the gathering of relatives and friends and fine Memorable some moments in the life such as landscape, one section of time, view at.

As people are increasingly diversified to the demand of photography and vedio recording technology, various photography and vedio recording camera lenses are by people Use and like.For example, to enable video-photographic equipment that there is broader visual-field space, " distortion figure Picture " arises at the historic moment.Fault image has the characteristics of focal length is short, visual field is big, has extensively in fully-directional visual system The market demand.

Fault image can reach the super large visual angle close to or greater than 180 °, so can be flutterred using fault image Larger range of scene is caught, therefore, fault image has huge potential using value.For example, by distortion figure Video monitoring system as being applied to some public arenas, using the mounting means of ceiling, then can make whole area The scene in domain is recorded.So people avoid the need for installing multiple monitoring cameras in different regions, to save Space, resource and use cost.Such as people always run into such case in daily life again, obviously feel View at the moment is very beautiful, how can not but be recorded with the video-photographic equipment in hand, and this is very big by one Partly cause is that the perspective capabilities of video-photographic equipment can not reach the scope to be seen of human eye.

Although fault image has the advantages that visual field is big, can reach even beyond the scope to be seen of human eye, But the visual angle of this super large of fault image is reached by sacrificing subject with its original form presentation 's.That is, there occurs distortion using the image captured by fault image.Fish eye images profile is rendered as circle Shape structure.Fault image can cause transparent effect strongly when being shot close to object, emphasize object Near big and far smaller contrast, makes taken the photograph picture have a kind of stirring appeal, therefore and enjoy shooting to like Person's likes.But, the image of this distortion is not in addition to it can strengthen artistic appeal, mostly by people institute Need.For example, the monitoring camera being seen everywhere in living now is set in some necessary places, can Help people's constraint daily behavior.Some monitoring records are it could even be possible to the vaild evidence of the identification that comes true.But It is that the picture of this deformation often influences the identification of some details.

Even if the image of distortion can give artistic appeal, many consumers are also desirable to the figure of these distortion Appearance as its script can be reduced to.Either it is used for commemorating or for being contrasted with fault image, All have very important significance is application value.Therefore, the alignment technique of fault image is deeply by research staff's Concern.

The premise of correcting fisheye image is exactly the contours extract of fish eye images.Conventional fish eye images profile is carried at present The method of taking has area statistics method, scan line approach method, region-growing method.This several method respectively has quality, but All there are some shortcomings, it is impossible to which entirely accurate positions the central coordinate of circle and radius of fish eye images, the scope of application There is limitation.

On the distortion correction of fish eye images, current method can mainly be summarized as 3D corrections and 2D corrections. The main method in the field is included based on the bearing calibration of spherical perspective projection model, based on quadratic surface perspective model Bearing calibration, fish eye images distortion correction method, the fish eye images plane based on geometric properties based on circle segmentation Bearing calibration etc..These above-mentioned methods, all respectively there is advantage and disadvantage, on computation complexity and calibration result, All do not comply fully with the real time correction requirement of HD video, all also existed in practical application it is certain away from From.

The real time correction of the fish eye images correction chart picture that obtains timely for consumer has great importance.Especially Correction for flake video image has great importance.Currently in the urgent need to a kind of high definition fish of real-time high-efficiency The bearing calibration of eye video.

The content of the invention

It is a primary object of the present invention to provide a distorted image correction method and its localization method, the wherein positioning Method can be accurately positioned the center of circle and the radius of fish eye images.

Another object of the present invention is to provide a distorted image correction method, wherein the method has correction rate Hurry up, the characteristics of calibration result is good.

Another object of the present invention is to provide a distorted image correction method, wherein the method can be used right Fish eye images are corrected.

Another object of the present invention is to provide a distorted image correction method, wherein the method is suitably used to right Fish eye images carry out real time correction.

Another object of the present invention is to provide a distorted image correction method, wherein the method can be used right Multichannel fish eye images are corrected.

Another object of the present invention is to provide a distorted image correction method, wherein the method has taken into full account fish The characteristics of eye image outline shows circular configuration simultaneously takes full advantage of the phase between multichannel flake video image Guan Xing, is adapted to realize on embedded the real time correction of multi-path high-definition fish eye images.

By following description, other advantages of the invention and feature will become apparent, it is possible to pass through The means and combination particularly pointed out in claims are accomplished.

According to an aspect of of the present present invention, the present invention provides a distorted image correction method, for fish eye images Correction, wherein the distorted image correction method includes：Following steps：

(1) correction parameter is determined；With

(2) according to the correction parameter, correcting algorithm is determined；

Wherein step (1) is comprised the following steps：

(11) position and the profile of multichannel fault image are determined；With

(12) correction factor of the fault image is determined；

Wherein step (11) is accurately positioned using bow and beam bearings to the position of the fault image and profile, To ensure the accurate and effective of the distorted image correction method.

According to one embodiment, the step (11) comprises the following steps：

(113) fault image is superimposed, to obtain a superimposed image；

(114) superimposed image described in linear compression, to obtain a normalized image；With

(115) position of the fault image is determined in the pixel value of each position according to the normalized image And profile.

According to one embodiment, before step (113), the step (11) comprises the following steps：

(112) fault image is filtered, to filter the noise of the fault image.

According to one embodiment, step (114) is comprised the following steps：

(1141) the max pixel value P of the superimposed image is obtained_maxWith minimum pixel value P_min；With

(1142) the max pixel value P according to the superimposed image_maxWith minimum pixel value P_minTo the superposition Image I_SCarry out linear compression；

Wherein linear compression is carried out using below equation：

Wherein, p_x,yIt is the later normalized image of linear compression in the pixel value at coordinate points (x, y) place, P_x,yIt is the pixel value at coordinate points (x, y) place on superimposed image.

According to one embodiment, step (114) is comprised the following steps：

(1141) the max pixel value P of the superimposed image is obtained_maxWith minimum pixel value P_min；With

(1142) the max pixel value P according to the superimposed image_maxWith minimum pixel value P_minTo the superposition Image I_SCarry out linear compression；

Wherein linear compression is carried out using below equation：

Wherein, p_x,yIt is the later normalized image of linear compression in the pixel value at coordinate points (x, y) place, P_x,yIt is the pixel value at coordinate points (x, y) place on superimposed image.

According to one embodiment, step (115) is comprised the following steps：

(1151) given threshold T_h；

(1152) record pixel in the normalized image and be more than or equal to threshold value T_hPoint；

(1153) it is more than or equal to according to pixel in the identified normalized image in step (1152) The point of the threshold value determines position and the profile of the fault image.

According to one embodiment, step (115) is comprised the following steps：

(1151) given threshold T_h；

(1152) record pixel in the normalized image and be more than or equal to threshold value T_hPoint；

(1153) it is more than or equal to according to pixel in the identified normalized image in step (1152) The point of the threshold value determines position and the profile of the fault image.

According to one embodiment, the threshold value T_hCan be obtained by below equation：

Wherein p_x,yBe the normalized image in the pixel value at coordinate points (x, y) place, W is the normalization figure The picture traverse of picture, H is the picture altitude of the normalized image.

According to one embodiment, the threshold value T_hCan be obtained by below equation：

Wherein p_x,yBe the normalized image in the pixel value at coordinate points (x, y) place, W is the normalization figure The picture traverse of picture, H is the picture altitude of the normalized image.

According to one embodiment, step (1152) is comprised the following steps：

(11521) normalized image is scanned from four direction；With

(11522) first for running into is separately recorded in during aforementioned four scanning direction to be more than or equal to The threshold value T_hPoint.

According to one embodiment, step (1152) is comprised the following steps：

(11521) normalized image is scanned from four direction；With

(11522) first for running into is separately recorded in during aforementioned four scanning direction to be more than or equal to The threshold value T_hPoint.

According to one embodiment, step (1) also includes a step：

(13) plane right-angle coordinate is set up；

Wherein step (1152) is further comprising the steps of：

(11523) it is big according to first run into the four direction scanning process in step (11522) In or equal to the threshold value T_hCoordinate value of the point in the plane right-angle coordinate be accurately positioned it is described abnormal Become the home position and imaging radius of image；

Wherein described four direction is included line by line from top to bottom, line by line from top to bottom, by column from left to right, by column From right to left, wherein respectively line by line from top to bottom, line by line from top to bottom, by column from left to right, by column from the right side First run into in left scanning process is more than or equal to the threshold value T_hPoint be respectively labeled as

Wherein step (1152) is further comprising the steps of：

(11524) two groups of vertical ranges and horizontal range of respective coordinates, wherein calculation are calculated respectively such as Shown in lower：

d₁=| y₁-y₂|

d₂=| x₃-x₄|

(11525) the imaging diameter d of the fault image is determined₃It is d₁And d₂In larger numerical value so that institute State the imaging radius R=d of fault image₃/2；With

(11536) home position of the fault image is determined, wherein the central coordinate of circle is (x_c,y_c), its In,

According to one embodiment, step (1) is further comprising the steps of：

(14) coordinate points of the profile point of the fault image in the plane right-angle coordinate are determined (x_il,y_i)；With

(15) the horizontal range l at the profile point range image center of the fault image is determined_ik；

Wherein

Wherein x_ilIt is the horizontal coordinate of the profile point of the i-th road fault image, y_iIt is the profile of the i-th road fault image The vertical coordinate of point, wherein l_ikFor the i-th road fault image vertical coordinate is y_kProfile point range image center Horizontal range.

According to one embodiment, step (12) comprises the following steps：

(121) angle point of fault image described in detection multichannel；

(122) angle point of superimposed image is detected；With

(123) according to the angle point and the angle point of the superimposed image per fault image described all the way, it is determined that respectively The correction factor α of road image_i。

Wherein, the angle point of fault image is sat in the flat square described in the multichannel for being detected in step (121) Coordinate in mark system is labeled as (x_ik,y_ik), which road video wherein i represents, and the angle point that k represents in the middle of video is compiled Number.

According to one embodiment, step (123) is comprised the following steps：

(1231) add up the abscissa of the angle point in the plane right-angle coordinate of each road fault image respectively Value, to obtain the angle point abscissa accumulated value X of fault image described in each road_i；

(1232) abscissa of the angle point of the cumulative superimposed image in the plane right-angle coordinate, with To the angle point abscissa accumulated value X of the superimposed image_M；

(1233) the correction factor α of the superimposed image is set_M；With

(1234) the correction factor α of each road fault image is calculated_i, wherein α_i=α_M·X_i/X_M。

According to one embodiment, the correction factor α of the superimposed image_MNumber range 0.7 and 1.3 it Between.

According to one embodiment, step (2) is further comprising the steps：

(21) according to the correction parameter obtained in step (1), determine that distortion correction formula is as follows：

According to one embodiment, before step (112), the step (11) also comprises the following steps：

(111) the gridiron pattern fault image of multichannel camera lens is gathered.

According to one embodiment, the distorted image correction method is further comprising the steps of：

(3) multichannel fault image is corrected according to the correcting algorithm.

According to one embodiment, step (3) is further comprising the steps：

(31) the distortion correction formula in step (21), generates distortion correction form.

According to one embodiment, step (3) is further comprising the steps：

(32) the correction form is applied to multi-path high-definition fault image under embedded system, to realize to institute State the real time correction of fault image.

According to a further aspect of the invention, the present invention also provides a fault image localization method, for flake The positioning of image, wherein the fault image localization method are comprised the following steps：

(113) multichannel fault image is superimposed, to obtain a superimposed image；

(114) superimposed image described in linear compression, to obtain a normalized image；With

(115) position of the fault image is determined in the pixel value of each position according to the normalized image And profile.

According to one embodiment, before step (113), the fault image localization method also includes as follows Step：

(112) fault image is filtered, to filter the noise of the fault image.

According to one embodiment, step (114) is comprised the following steps：

(1141) the max pixel value P of the superimposed image is obtained_maxWith minimum pixel value P_min；With

(1142) the max pixel value P according to the superimposed image_maxWith minimum pixel value P_minTo the superposition Image I_SCarry out linear compression；

Wherein linear compression is carried out using below equation：

Wherein, p_x,yIt is the later normalized image of linear compression in the pixel value at coordinate points (x, y) place, P_x,yIt is the pixel value at coordinate points (x, y) place on superimposed image.

According to one embodiment, step (114) is comprised the following steps：

(1141) the max pixel value P of the superimposed image is obtained_maxWith minimum pixel value P_min；With

(1142) the max pixel value P according to the superimposed image_maxWith minimum pixel value P_minTo the superposition Image I_SCarry out linear compression；

Wherein linear compression is carried out using below equation：

Wherein, p_x,yIt is the later normalized image of linear compression in the pixel value at coordinate points (x, y) place, P_x,yIt is the pixel value at coordinate points (x, y) place on superimposed image.

According to one embodiment, step (115) is comprised the following steps：

(1151) given threshold T_h；

(1152) record pixel in the normalized image and be more than or equal to threshold value T_hPoint；

(1153) it is more than or equal to according to pixel in the identified normalized image in step (1152) The point of the threshold value determines position and the profile of the fault image.

According to one embodiment, step (115) is comprised the following steps：

(1151) given threshold T_h；

(1152) record pixel in the normalized image and be more than or equal to threshold value T_hPoint；

(1153) it is more than or equal to according to pixel in the identified normalized image in step (1152) The point of the threshold value determines position and the profile of the fault image.

According to one embodiment, the threshold value T_hCan be obtained by below equation：

Wherein p_x,yBe the normalized image in the pixel value at coordinate points (x, y) place, W is the normalization figure The picture traverse of picture, H is the picture altitude of the normalized image.

According to one embodiment, the threshold value T_hCan be obtained by below equation：

Wherein p_x,yBe the normalized image in the pixel value at coordinate points (x, y) place, W is the normalization figure The picture traverse of picture, H is the picture altitude of the normalized image.

According to one embodiment, step (1152) is comprised the following steps：

(11521) normalized image is scanned from four direction；With

(11522) first for running into is separately recorded in during aforementioned four scanning direction to be more than or equal to The threshold value T_hPoint.

According to one embodiment, step (1152) is comprised the following steps：

(11521) normalized image is scanned from four direction；With

(11522) first for running into is separately recorded in during aforementioned four scanning direction to be more than or equal to The threshold value T_hPoint.

According to one embodiment, the fault image localization method also includes a step：

(13) plane right-angle coordinate is set up；

Wherein step (1152) is further comprising the steps of：

(11523) it is big according to first run into the four direction scanning process in step (11522) In or equal to the threshold value T_hCoordinate value of the point in the plane right-angle coordinate be accurately positioned it is described abnormal Become the home position and imaging radius of image；

Wherein described four direction is included line by line from top to bottom, line by line from top to bottom, by column from left to right, by column From right to left, wherein respectively line by line from top to bottom, line by line from top to bottom, by column from left to right, by column from the right side First run into in left scanning process is more than or equal to the threshold value T_hPoint be respectively labeled as

Wherein step (1152) is further comprising the steps of：

(11524) two groups of vertical ranges and horizontal range of respective coordinates, wherein calculation are calculated respectively such as Shown in lower：

d₁=| y₁-y₂|

d₂=| x₃-x₄|

(11525) the imaging diameter d of the fault image is determined₃It is d₁And d₂In larger numerical value so that institute State the imaging radius R=d of fault image₃/2；With

(11536) home position of the fault image is determined, wherein the central coordinate of circle is (x_c,y_c), its In,

According to one embodiment, the fault image localization method is further comprising the steps of：

(14) coordinate points of the profile point of the fault image in the plane right-angle coordinate are determined (x_il,y_i)；With

(15) the horizontal range l at the profile point range image center of the fault image is determined_ik；

Wherein

Wherein x_ilIt is the horizontal coordinate of the profile point of the i-th road fault image, y_iIt is the profile of the i-th road fault image The vertical coordinate of point, wherein l_ikFor the i-th road fault image vertical coordinate is y_kProfile point range image center Horizontal range.

According to one embodiment, before step (112), the fault image localization method also includes as follows Step：

(111) the gridiron pattern fault image of multichannel camera lens is gathered.

By the understanding to subsequent description and accompanying drawing, further aim of the present invention and advantage will be able to abundant body It is existing.

These and other objects of the invention, feature and advantage, by following detailed descriptions, accompanying drawing and right It is required that being fully demonstrated.

Brief description of the drawings

Fig. 1 is the filtering used in a distorted image correction method according to a preferred embodiment of the present invention Template schematic diagram.

Fig. 2 is the distorted image correction method schematic diagram of above preferred embodiment of the invention.

Fig. 3 illustrates a determination distortion of the distorted image correction method of above preferred embodiment of the invention The step of image outline.

Fig. 4 illustrates a determination correction of the distorted image correction method of above preferred embodiment of the invention The step of factor.

Fig. 5 is illustrated according to the distorted image correction method schematic diagram of above preferred embodiment of the invention.

Specific embodiment

Hereinafter describe for disclosing the present invention so that those skilled in the art can realize the present invention.In below describing Preferred embodiment be only used as citing, it may occur to persons skilled in the art that other obvious modifications.With The general principle of the invention defined in lower description can apply to other embodiments, deformation program, improvement side Case, equivalent and the other technologies scheme without departing from the spirit and scope of the present invention.

Fig. 1 to Fig. 4 of accompanying drawing is illustrated according to the distorted image correction side of a preferred embodiment of the present invention Method.The distorted image correction method can be applied to the distortion correction of fish eye images, but be not limited to fish-eye image The distortion correction of picture.Those skilled in the art are it should be appreciated that any to meet fish eye images profile rounded abnormal The correction of the fault image of change is all suitable for the distorted image correction method.This preferred embodiment is with multichannel fault image Distorted image correction as a example by distorted image correction method of the invention is described in detail.

As shown in Fig. 2 of accompanying drawing, the distorted image correction method is comprised the following steps：

(1) correction parameter is determined；

(2) according to the correction parameter, correcting algorithm is determined；With

(3) real time correction is carried out to multichannel fault image according to the correcting algorithm.

Wherein correction parameter is determined according to the fault image described in step (1).Specifically, The step (1) comprises the following steps：

(11) profile per fault image described all the way is determined；With

(12) the correction factor α per fault image described all the way is determined_i。

More specifically, the step (11) comprises the following steps：

(111) the gridiron pattern fault image of multichannel camera lens is gathered；

(112) fault image is filtered, to filter the noise on each road fault image, so as to prevent Correction of the noise to the fault image is impacted；

(113) fault image that superposition is filtered through the step (112), to obtain a superimposed image I_S；

(114) superimposed image I described in linear compression_S, to obtain a normalized image I_M；With

(115) according to normalized image I_MIn the pixel value p of each position_x,y, it is accurately positioned the distortion figure The position of picture and profile.

The fault image for wherein being gathered by step (111) is deposited due to the influence by some factors In some noises, can be to fault image profile fixed output quota life interference really, so needing to the fault image Filtered, to reduce the influence that the noise determines to the fault image profile.Filtering employed in it Template is as shown in Figure 1.

Those skilled in the art for no noise or noise it should be appreciated that be as low as not enough to distortion figure The correction of picture produces the fault image of influence, then need not be corrected.If that is, passing through step (111) The fault image for being gathered does not have noise or noise to be as low as not enough to produce shadow to the correction of the figure that distorts Ring, then step (112) is not required.So, can directly to the fault image in step (113) It is overlapped.That is, step (113) is changed into being superimposed the fault image, to obtain a superimposed image I_S。

In order to more accurately be positioned to fault image described in each road, to the distortion in the step (113) Image is superimposed, and the error of each road fault image position is determined to offset.For alternatively, if Fault image described in each road is positioned respectively, various environmental factors or human factor is inevitably produced It is not accurate enough that the error for causing, wherein these errors not only result in fault image positioning, and can cause It is possible to be misplaced because respective error is different between each road fault image, after further resulting in correction Picture quality cannot be protected.So, each road fault image is carried out by image superimposing method unify positioning, Quality after being conducive to guarantee image to be corrected.

Wherein step (114) is comprised the following steps：

(1141) the superimposed image I is obtained_SMax pixel value P_maxWith minimum pixel value P_min；With

(1142) according to the superimposed image I_SMax pixel value P_maxWith minimum pixel value P_minTo described folded Plus image I_SCarry out linear compression；

Wherein linear compression is carried out using formula 1：

Formula 1

Wherein, p_x,yIt is the later normalized image I of linear compression_MIn the pixel value at coordinate points (x, y) place, P_x,yIt is superimposed image I_SThe pixel value at upper coordinate points (x, y) place.

In addition, it is noted that this preferred embodiment is indeformable using fault image shooting image center section And the characteristics of peripheral outline rounded flexural distortion, first to the center of circle of circle where circle fault image peripheral outline Positioned, and then fish eye images profile is accurately positioned.This center of circle positioning mode had not only facilitated but also accurate, The characteristics of making the distorted image correction method have simple, efficient.

Specifically, the step (115) is comprised the following steps：

(1151) given threshold T_h；

(1152) the normalized image I is recorded_MUpper pixel is more than or equal to threshold value T_hPoint；

(1153) according to the identified normalized image I in step (1152)_MUpper pixel is more than or waits In threshold value T_hPoint determine position and the profile of the fault image.

Further, the step (115) is for determining the method for the fault image position and profile for determine at 4 points Position method.Specifically, the step (1152) is comprised the following steps：

(11521) to the normalized image I from four direction_MIt is scanned；With

(11522) first for running into is separately recorded in during aforementioned four scanning direction to be more than or equal to The threshold value T_hPoint.

More specifically, the four direction be respectively line by line from top to bottom, line by line from top to bottom, by column from a left side to It is right and by column from right to left.

Wherein described threshold value T_hObtained by equation 2 below：

Formula 2

Wherein p_x,yIt is the normalized image I_MIn the pixel value at coordinate points (x, y) place, W is the normalization The picture traverse of image, H is the picture altitude of the normalized image.

To make the distorted image correction method quicker, accurate and effective, the distorted image correction method should Step (1) also includes a step：

(13) plane right-angle coordinate is set up.

It is noted that between the step (11), the step (12) and the step (13) not first The difference of order afterwards, the sequencing between three can be exchanged and unrestricted.

The plane right-angle coordinate set up in step (13) makes every on image in the distorted image correction method Can a little be determined with specific coordinate value, so assist in the coordinate system with the fault image school The related relative position relation of correction method.

On the other hand, because the every bit in the plane right-angle coordinate can enter rower by specific numerical value It is fixed, so the convenient specific mathematical relationship using geometric figure is accurately determined to geometric figure.In the present invention The preferred embodiment in, because fish eye images profile is rendered as circular configuration, this preferred embodiment is using circle Mathematical relationship is accurately positioned to fish eye images profile, so that the distorted image correction is more accurate.It is another Aspect, conveniently carries out mathematical computations.

As shown in Fig. 2 in superimposed image I_SThe plane right-angle coordinate is set up in the plane at place, wherein should Plane right-angle coordinate is made up of an orthogonal X-axis and a Y-axis, and wherein the X-axis and the Y-axis intersect In an origin O, the coordinate of the wherein coordinate points in the coordinate system is denoted as (x, y).

It is noted that the plane right-angle coordinate is set up for the convenience for calculating and demarcating, to the present invention Not tangible restriction effect.That is, no matter the coordinate system is based upon superimposed image I_SPlace Plane in where, do not affect calibration result of the distorted image correction method to fault image.Namely Say, each coordinate points (x, y) is to play relative marked effect, and the concrete numerical value of wherein x and y is not In the presence of absolute meaning.

Accordingly, the step (1152) is comprised the following steps：

(11523) it is big according to first run into the four direction scanning process in step (11522) In or equal to the threshold value T_hCoordinate value of the point in the plane right-angle coordinate be accurately positioned it is described abnormal Become the home position and imaging radius of image.

Specifically, scan from right to left from left to right, by column from top to bottom, by column from top to bottom, line by line line by line During find meet require pixel be respectively labeled as

Two groups of vertical ranges and horizontal range of respective coordinates, its calculation such as formula 3, formula are calculated respectively Shown in 4：

d₁=| y₁-y₂| formula 3

d₂=| x₃-x₄| formula 4

Selection d₁And d₂In larger numerical value as fault image imaging diameter d₃, then obtain fault image into As radius R, central coordinate of circle is (x_c,y_c)。R、x_c、y_cCounted by formula 5, formula 6, formula 7 respectively Obtain：

R=d₃/ 2 formula 5

Formula 6

Formula 7

The step (1) also includes step：

(14) coordinate points of the profile point of the fault image in the plane right-angle coordinate are determined (x_il,y_i)；With

(15) the horizontal range l at the profile point range image center of the fault image is determined_ik。

The profile of wherein described fault image refers to the peripheral outline of the fault image.Wherein in step (14) The coordinate value of coordinate points of the profile point of the fault image in the plane right-angle coordinate is by following public affairs Formula is determined：

Formula 8

Wherein x_ilIt is the horizontal coordinate of the profile point of the i-th road fault image, y_iIt is the profile point of the i-th road fault image Vertical coordinate.

The horizontal range l at the profile point range image center of fault image described in step (15)_ikBy following public affairs Formula is determined：

Formula 9

Wherein l_ikFor the i-th road fault image vertical coordinate is y_kProfile point range image center horizontal range.

It is noted that according to the distorted image correction method, step of presently preferred embodiment of the invention (15) the horizontal range l at the profile point range image center of fault image described in_ikSat by the flat square Mark is, and the mathematical relationship formula that make use of several picture profile existing is calculated so that l_ikNumerical value precisely, So as to ensure that the accuracy and accuracy of the distorted image correction method.But those skilled in the art should be able to It is enough to understand, this be only to example of the invention, it is and unrestricted.

The step (12) comprises the following steps：

(121) angle point of fault image described in detection multichannel；

(122) detection superimposed image I_SAngle point；With

(123) according to the angle point and the superimposed image I per fault image described all the way_SAngle point, it is determined that The correction factor α of each road image_i。

Specifically, the angle point of fault image is in the flat square described in the multichannel for being detected in step (121) Coordinate in coordinate system is labeled as (x_ik,y_ik), which road video wherein i represents, and k represents the angle point in the middle of video Numbering.

Step (123) is comprised the following steps：

(1231) add up the abscissa of the angle point in the plane right-angle coordinate of each road fault image respectively Value, to obtain the angle point abscissa accumulated value X of fault image described in each road_i；

(1232) add up the superimposed image I_SAbscissa of the angle point in the plane right-angle coordinate, with Obtain the superimposed image I_SAngle point abscissa accumulated value X_M；

(1233) the superimposed image I is set_SCorrection factor α_M(scope is between 0.7 and 1.3)；With

(1234) according to X_i、X_MAnd α_MCalculate the correction factor α of each road fault image_i, wherein α_i=α_M·X_i/X_M。

Wherein, in step (1231) be used for add up each road fault image angle point in the plane right-angle coordinate In abscissa value formula as shown in Equation 10：

Formula 10

Wherein, x_ikK-th abscissa size of angle point under the i-th road gridiron pattern fault image is represented, K represents each The angle point number altogether of fault image described in road.

It is noted that there is no the differentiation of sequencing between the step (121) and (122), both it Between sequencing can exchange.Those skilled in the art are it should be appreciated that the step (121) and (122) Can also carry out simultaneously.That is, according to presently preferred embodiment of the invention, the step (121) and (122) There is no the difference on any sequencing.

Step (2) is further comprising the steps：

(21) according to the correction parameter obtained in step (1), distortion correction formula is determined：

Formula 11

Wherein, a_iIt is the radius of the i-th road flake video image major axis；b_i=1,2,3 ..., H；H is abnormal Become image radius wide, x_ilIt is the horizontal coordinate of the profile of the i-th road fault image, x_cIt is the fault image The horizontal coordinate at center, l_iFor the i-th road fault image profile horizontal coordinate to the center of the fault image Horizontal coordinate distance, α_iIt is the correction factor of the i-th road fault image, reflects the size of correction amplitude.

Step (3) is further comprising the steps：

(31) according to formula 11, distortion correction form is generated；With

(32) the correction form is applied to multi-path high-definition fault image under embedded system, to realize to institute State the real time correction of fault image.

It is noted that in the present invention the step of distorted image correction method in used in numbering used The Arabic numerals such as 1,2,3,4,5 be only to play marked effect, and do not differentiate between the effect of precedence. Those skilled in the art are not it should be appreciated that violating each step in itself in the case of logical order, this is abnormal It is that no precedence is distinguished to become the step in method for correcting image.Certainly, those skilled in the art should be able to It is enough to understand, some subsequent steps need by above the step of premised in the case of, these steps are that have first The differentiation of order afterwards.And for those mutually premised on the step of, as long as the purpose of the present invention can be realized, It sequentially can be what is be exchanged with each other.

In order to the present invention is described in more detail, below by taking the correction of flake video image as an example, to the fault image school Correction method is explained in further detail.

The distorted image correction method gathers the chessboard table images of multichannel fault image, and these chessboard table images are entered Row low-pass filtering operation, filters the high-frequency noise on image, eliminates related influence.The Filtering Template for being used As shown in Figure 1.

Pretreated multichannel fish eye images are overlapped, superimposed image I is obtained_S.Traversal superimposed image I_S, Obtain maximum P_maxWith minimum value P_min.By P_maxAnd P_min, by superimposed image I_SCarry out linear compression, Obtain normalized image I_M, I_MThe pixel value range of image is between 0 to 255.Linear compression uses formula 1 is carried out.

Given threshold T_hIt is the average of superimposed image, its calculation is as shown in Equation 2.

To normalized image I_MThe scanning pixel-by-pixel of four direction is carried out, when each scanning direction, record the One is more than or equal to threshold value T_hPixel coordinate position.Four scanning directions be respectively line by line on to Under, line by line from top to bottom, by column from left to right, by column from right to left.Found in scanning process and meet what is required Pixel is respectively labeled as

Two groups of vertical ranges and horizontal range of respective coordinates, its calculation such as formula 3, formula are calculated respectively Shown in 4.

Selection d₁And d₂In larger numerical value as fish eye images imaging diameter d₃, then obtain fish eye images into As radius R, central coordinate of circle is (x_c,y_c)。R、x_c、y_cCounted by formula 5, formula 6, formula 7 respectively Obtain.

Coordinate (the x of the profile point of each road flake video image is obtained by formula 8_il,y_i)。

The horizontal range l at the profile point range image center of each road flake video image is obtained by formula 9_ik。

By Corner Detection Algorithm, the angle point (x in the middle of each road chessboard table images is detected_ik,y_ik), wherein i is represented Which road video, k represents the angle point numbering in the middle of video.

Abscissa to each road flake video angle point adds up, and the angle point abscissa for obtaining each road flake video tires out It is value added.

Detection superimposed image I_SAngle point, and add up all angle points abscissa, obtain X_M。

Set the correction factor α of superimposed image_M(scope is between 07 and 13), then can obtain other each roads The correction factor α of video_i=α_M·X_i/X_M。

By various parameters resulting before, the distortion correction formula 11 of each road flake video is comprehensively obtained.

Using formula 11, then each road flake video, generation and the one-to-one distortion correction table of pixel can be directed to Lattice, meet the real time correction demand of multi-path high-definition flake video under embedded system.

It should be understood by those skilled in the art that the embodiments of the invention shown in foregoing description and accompanying drawing are only used as Illustrate and be not intended to limit the present invention.The purpose of the present invention completely and is effectively realized.Function of the invention and Structural principle shows and illustrates in embodiment, under without departing from the principle, embodiments of the present invention Can there are any deformation or modification.

Claims (34)

1. a distorted image correction method, for the correction of fish eye images, it is characterised in that including following Step：

(1) correction parameter is determined；With

(2) according to the correction parameter, correcting algorithm is determined；

Wherein step (1) is comprised the following steps：

(11) position and the profile of multichannel fault image are determined；With

(12) correction factor of the fault image is determined；

Wherein step (11) is accurately positioned using bow and beam bearings to the position of the fault image and profile, To ensure the accurate and effective of the distorted image correction method.

2. distorted image correction method according to claim 1, wherein the step (11) is including such as Lower step：

(113) fault image is superimposed, to obtain a superimposed image；

(114) superimposed image described in linear compression, to obtain a normalized image；With

(115) position of the fault image is determined in the pixel value of each position according to the normalized image And profile.

3. distorted image correction method according to claim 2, wherein before step (113), institute Step (11) is stated to comprise the following steps：

(112) fault image is filtered, to filter the noise of the fault image.

4. distorted image correction method according to claim 2, wherein step (114) includes following step Suddenly：

(1141) the max pixel value P of the superimposed image is obtained_maxWith minimum pixel value P_min；With

(1142) the max pixel value P according to the superimposed image_maxWith minimum pixel value P_minTo the superposition Image I_SCarry out linear compression；

Wherein linear compression is carried out using below equation：

$p_{x,y}=\frac{P_{x,y}-P_{min}}{P_{\max }-P_{\min }}\·255$

Wherein, p_x,yIt is the later normalized image of linear compression in the pixel value at coordinate points (x, y) place, P_x,yIt is the pixel value at coordinate points (x, y) place on superimposed image.

5. distorted image correction method according to claim 3, wherein step (114) includes following step Suddenly：

(1141) the max pixel value P of the superimposed image is obtained_maxWith minimum pixel value P_min；With

(1142) the max pixel value P according to the superimposed image_maxWith minimum pixel value P_minTo the superposition Image I_SCarry out linear compression；

Wherein linear compression is carried out using below equation：

$p_{x,y}=\frac{P_{x,y}-P_{min}}{P_{\max }-P_{\min }}\·255$

Wherein, p_x,yIt is the later normalized image of linear compression in the pixel value at coordinate points (x, y) place, P_x,yIt is the pixel value at coordinate points (x, y) place on superimposed image.

6. distorted image correction method according to claim 4, wherein step (115) includes following step Suddenly：

(1151) given threshold T_h；

(1152) record pixel in the normalized image and be more than or equal to threshold value T_hPoint；

(1153) it is more than or equal to according to pixel in the identified normalized image in step (1152) The point of the threshold value determines position and the profile of the fault image.

7. distorted image correction method according to claim 5, wherein step (115) includes following step Suddenly：

(1151) given threshold T_h；

(1152) record pixel in the normalized image and be more than or equal to threshold value T_hPoint；

(1153) it is more than or equal to according to pixel in the identified normalized image in step (1152) The point of the threshold value determines position and the profile of the fault image.

8. distorted image correction method according to claim 6, wherein the threshold value T_hCan be by following Formula is obtained：

$T_h=\frac{{\mathrm{\Σp}}_{x,y}}{W\·H}$

Wherein p_x,yBe the normalized image in the pixel value at coordinate points (x, y) place, W is the normalization figure The picture traverse of picture, H is the picture altitude of the normalized image.

9. distorted image correction method according to claim 7, wherein the threshold value T_hCan be by following Formula is obtained：

$T_h=\frac{{\mathrm{\Σp}}_{x,y}}{W\·H}$

Wherein p_x,yBe the normalized image in the pixel value at coordinate points (x, y) place, W is the normalization figure The picture traverse of picture, H is the picture altitude of the normalized image.

10. distorted image correction method according to claim 8, wherein step (1152) is including following Step：

(11521) normalized image is scanned from four direction；With

(11522) first for running into is separately recorded in during aforementioned four scanning direction to be more than or equal to The threshold value T_hPoint.

11. distorted image correction methods according to claim 9, wherein step (1152) is including following Step：

(11521) normalized image is scanned from four direction；With

(11522) first for running into is separately recorded in during aforementioned four scanning direction to be more than or equal to The threshold value T_hPoint.

The 12. distorted image correction method according to any one in claim 1~11, wherein step (1) Also include a step：

(13) plane right-angle coordinate is set up；

Wherein step (1152) is further comprising the steps of：

(11523) it is big according to first run into the four direction scanning process in step (11522) In or equal to the threshold value T_hCoordinate value of the point in the plane right-angle coordinate be accurately positioned it is described abnormal Become the home position and imaging radius of image；

Wherein described four direction is included line by line from top to bottom, line by line from top to bottom, by column from left to right, by column From right to left, wherein respectively line by line from top to bottom, line by line from top to bottom, by column from left to right, by column from the right side First run into in left scanning process is more than or equal to the threshold value T_hPoint be respectively labeled as

Wherein step (1152) is further comprising the steps of：

(11524) two groups of vertical ranges and horizontal range of respective coordinates, wherein calculation are calculated respectively such as Shown in lower：

d₁=| y₁-y₂|

d₂=| x₃-x₄|

(11525) the imaging diameter d of the fault image is determined₃It is d₁And d₂In larger numerical value so that institute State the imaging radius R=d of fault image₃/2；With

(11536) home position of the fault image is determined, wherein the central coordinate of circle is (x_c,y_c), its In,

$x_c=\frac{x_3+x_4}{2};$

$y_c=\frac{y_1+y_2}{2}.$

13. distorted image correction methods according to claim 12, wherein step (1) also include following Step：

(14) coordinate points of the profile point of the fault image in the plane right-angle coordinate are determined (x_il,y_i)；With

(15) the horizontal range l at the profile point range image center of the fault image is determined_ik；

Wherein $x_{il}=x_c-R\·\cos \arcsin \left(\frac{y_c-y_i}{R}\right),$ $l_{ik}=R\·\cos \arcsin \frac{y_c-y_k}{R},$

Wherein x_ilIt is the horizontal coordinate of the profile point of the i-th road fault image, y_iIt is the profile of the i-th road fault image The vertical coordinate of point, wherein l_ikFor the i-th road fault image vertical coordinate is y_kProfile point range image center Horizontal range.

14. distorted image correction methods according to claim 13, wherein step (12) include following step Suddenly：

(121) angle point of fault image described in detection multichannel；

(122) angle point of superimposed image is detected；With

(123) according to the angle point and the angle point of the superimposed image per fault image described all the way, it is determined that respectively The correction factor α of road image_i。

Wherein, the angle point of fault image is sat in the flat square described in the multichannel for being detected in step (121) Coordinate in mark system is labeled as (x_ik,y_ik), which road video wherein i represents, and the angle point that k represents in the middle of video is compiled Number.

15. distorted image correction methods according to claim 14, wherein step (123) is including following Step：

(1231) add up the abscissa of the angle point in the plane right-angle coordinate of each road fault image respectively Value, to obtain the angle point abscissa accumulated value X of fault image described in each road_i；

(1232) abscissa of the angle point of the cumulative superimposed image in the plane right-angle coordinate, with To the angle point abscissa accumulated value X of the superimposed image_M；

(1233) the correction factor α of the superimposed image is set_M；With

(1234) the correction factor α of each road fault image is calculated_i, wherein α_i=α_M·X_i/X_M。

16. distorted image correction methods according to claim 15, wherein the correction of the superimposed image Factor-alpha_MNumber range between 0.7 and 1.3.

17. distorted image correction methods according to claim 16, wherein step (2) are further included Following steps：

(21) according to the correction parameter obtained in step (1), determine that distortion correction formula is as follows：

$\frac{{y_i}^2}{{a_i}^2}+\frac{{(x_{i1}+{\mathrm{\α}}_i(x_i-x_c+l_i\left)\right)}^2}{{b_i}^2}=1$

18. distorted image correction methods according to claim 17, wherein before step (112), The step (11) also comprises the following steps：

(111) the gridiron pattern fault image of multichannel camera lens is gathered.

19. distorted image correction methods according to claim 18, wherein the distorted image correction side Method is further comprising the steps of：

(3) multichannel fault image is corrected according to the correcting algorithm.

20. distorted image correction methods according to claim 19, wherein step (3) are further included Following steps：

(31) the distortion correction formula in step (21), generates distortion correction form.

21. distorted image correction methods according to claim 20, wherein step (3) are further included Following steps：

(32) the correction form is applied to multi-path high-definition fault image under embedded system, to realize to institute State the real time correction of fault image.

22. 1 fault image localization methods, for the positioning of fish eye images, it is characterised in that including following Step：

(113) multichannel fault image is superimposed, to obtain a superimposed image；

(114) superimposed image described in linear compression, to obtain a normalized image；With

(115) position of the fault image is determined in the pixel value of each position according to the normalized image And profile.

23. fault image localization methods according to claim 22, wherein before step (113), The fault image localization method also comprises the following steps：

(112) fault image is filtered, to filter the noise of the fault image.

24. fault image localization methods according to claim 22, wherein step (114) is including following Step：

(1141) the max pixel value P of the superimposed image is obtained_maxWith minimum pixel value P_min；With

(1142) the max pixel value P according to the superimposed image_maxWith minimum pixel value P_minTo the superposition Image I_SCarry out linear compression；

Wherein linear compression is carried out using below equation：

$p_{x,y}=\frac{P_{x,y}-P_{min}}{P_{\max }-P_{\min }}\·255$

Wherein, p_x,yIt is the later normalized image of linear compression in the pixel value at coordinate points (x, y) place, P_x,yIt is the pixel value at coordinate points (x, y) place on superimposed image.

25. fault image localization methods according to claim 23, wherein step (114) is including following Step：

(1141) the max pixel value P of the superimposed image is obtained_maxWith minimum pixel value P_min；With

(1142) the max pixel value P according to the superimposed image_maxWith minimum pixel value P_minTo the superposition Image I_SCarry out linear compression；

Wherein linear compression is carried out using below equation：

$p_{x,y}=\frac{P_{x,y}-P_{min}}{P_{\max }-P_{\min }}\·255$

Wherein, p_x,yIt is the later normalized image of linear compression in the pixel value at coordinate points (x, y) place, P_x,yIt is the pixel value at coordinate points (x, y) place on superimposed image.

26. fault image localization methods according to claim 24, wherein step (115) is including following Step：

(1151) given threshold T_h；

(1152) record pixel in the normalized image and be more than or equal to threshold value T_hPoint；

(1153) it is more than or equal to according to pixel in the identified normalized image in step (1152) The point of the threshold value determines position and the profile of the fault image.

27. fault image localization methods according to claim 25, wherein step (115) is including following Step：

(1151) given threshold T_h；

(1152) record pixel in the normalized image and be more than or equal to threshold value T_hPoint；

(1153) it is more than or equal to according to pixel in the identified normalized image in step (1152) The point of the threshold value determines position and the profile of the fault image.

28. fault image localization methods according to claim 26, wherein the threshold value T_hCan by with Lower formula is obtained：

$T_h=\frac{{\mathrm{\Σp}}_{x,y}}{W\·H}$

Wherein p_x,yBe the normalized image in the pixel value at coordinate points (x, y) place, W is the normalization figure The picture traverse of picture, H is the picture altitude of the normalized image.

29. fault image localization methods according to claim 27, wherein the threshold value T_hCan by with Lower formula is obtained：

$T_h=\frac{{\mathrm{\Σp}}_{x,y}}{W\·H}$

Wherein p_x,yBe the normalized image in the pixel value at coordinate points (x, y) place, W is the normalization figure The picture traverse of picture, H is the picture altitude of the normalized image.

30. fault image localization methods according to claim 28, wherein step (1152) is including following Step：

(11521) normalized image is scanned from four direction；With

(11522) first for running into is separately recorded in during aforementioned four scanning direction to be more than or equal to The threshold value T_hPoint.

31. fault image localization methods according to claim 29, wherein step (1152) is including following Step：

(11521) normalized image is scanned from four direction；With

(11522) first for running into is separately recorded in during aforementioned four scanning direction to be more than or equal to The threshold value T_hPoint.

The 32. fault image localization method according to any one in claim 22~31, wherein described abnormal Becoming image position method also includes a step：

(13) plane right-angle coordinate is set up；

Wherein step (1152) is further comprising the steps of：

(11523) it is big according to first run into the four direction scanning process in step (11522) In or equal to the threshold value T_hCoordinate value of the point in the plane right-angle coordinate be accurately positioned it is described abnormal Become the home position and imaging radius of image；

Wherein described four direction is included line by line from top to bottom, line by line from top to bottom, by column from left to right, by column From right to left, wherein respectively line by line from top to bottom, line by line from top to bottom, by column from left to right, by column from the right side First run into in left scanning process is more than or equal to the threshold value T_hPoint be respectively labeled as

Wherein step (1152) is further comprising the steps of：

(11524) two groups of vertical ranges and horizontal range of respective coordinates, wherein calculation are calculated respectively such as Shown in lower：

d₁=| y₁-y₂|

d₂=| x₃-x₄|

(11525) the imaging diameter d of the fault image is determined₃It is d₁And d₂In larger numerical value so that institute State the imaging radius R=d of fault image₃/2；With

(11536) home position of the fault image is determined, wherein the central coordinate of circle is (x_c,y_c), its In,

$x_c=\frac{x_3+x_4}{2};$

$y_c=\frac{y_1+y_2}{2}.$

33. fault image localization methods according to claim 32, wherein the fault image localization method It is further comprising the steps of：

(14) coordinate points of the profile point of the fault image in the plane right-angle coordinate are determined (x_il,y_i)；With

(15) the horizontal range l at the profile point range image center of the fault image is determined_ik；

Wherein $x_{il}=x_c-R\·\cos \arcsin \left(\frac{y_c-y_i}{R}\right),$ $l_{ik}=R\·\cos \arcsin \frac{y_c-y_k}{R},$

Wherein x_ilIt is the horizontal coordinate of the profile point of the i-th road fault image, y_iIt is the profile of the i-th road fault image The vertical coordinate of point, wherein l_ikFor the i-th road fault image vertical coordinate is y_kProfile point range image center Horizontal range.

34. fault image localization methods according to claim 33, wherein before step (112), The fault image localization method also comprises the following steps：

(111) the gridiron pattern fault image of multichannel camera lens is gathered.