Summary of the invention
Goal of the invention: in order to overcome the deficiencies in the prior art, the invention provides a kind of vehicle-mounted viewing system camera parameter calibration method, by using scaling board, and calibrating parameters grouping, by entering the method that every group of parameter demarcated, has been solved to vehicle-mounted viewing system in prior art because the problem of Image Mosaics precision is looked around in positioning error impact.
Technical scheme: a kind of vehicle-mounted viewing system camera parameter calibration method, it is characterized in that, when vehicle-mounted viewing system camera inner parameter and calibrating external parameters, adopt three-dimensional symbol mark to demarcate.
Three-dimensional symbol mark is positioned on three-dimensional correction mark, and the three-dimensional scaling coordinate system coordinate of this sign mark is known.
In viewing system, the FOV (Field of View) of each camera can clearly be seen 3 above described three-dimensional symbol marks.
Comprise vehicle-mounted viewing system and calibration mark thing; Described vehicle-mounted viewing system comprises 2-4 platform camera, and camera is placed in middle part all around, vehicle top;
The method comprises the following steps:
1) put calibration mark thing: demarcate on the ground a region that can hold vehicle, vehicle to be corrected is parked in this rectangular area;
Described three-dimensional correction mark comprises some unique points that can identify from image and that three-dimensional coordinate is known, such as angle point, and the unique points such as round dot;
2) park vehicle to be corrected acquisition correction view data:
Sail vehicle to be corrected into step 1) in described region, carry out artificial or carry out automated graphics data acquisition by external trigger signal.
3, a kind of vehicle-mounted viewing system camera parameter calibration method as claimed in claim 1: it is characterized in that, to every camera calibration step of vehicle-mounted viewing system camera:
3.1) obtain calibration mark thing unique point: by feature point extraction way extract minutiae from each image, and mate with known unique point three-dimensional coordinate;
3.2) take step 3.1) two dimensional image coordinate and the corresponding known three-dimensional coordinate of the unique point obtained be measurement data, utilizes the inside and outside parameter of method of value solving calibration for cameras; By parameter grouping, by demarcating; While demarcating wherein one group of parameter, other parameters remain unchanged;
3.3) utilize step 3.2) the camera inside and outside parameter that obtains carries out projective transformation to the mark unique point in three dimensions, projected on original image, calculates projection coordinate and the residual error of measuring coordinate;
3.4) default cycle index and residual error threshold value, judge whether to exceed cycle index threshold value or determining step 3.3) whether the residual error that obtains lower than predetermined threshold value;
If do not exceeded cycle index threshold value and residual error, be not less than predetermined threshold value, circulation step 3.1) to step 3.3);
As exceed cycle index threshold value or residual error lower than predetermined threshold value, judge whether residual error is greater than predetermined threshold value; As residual error is greater than predetermined threshold value, system provides the information of again demarcating, repeating step 3.1) to step 3.3); As residual threshold value is not more than predetermined threshold value, demarcate and finish, store and demarcate achievement.
Beneficial effect:
1, adopt the scaling board perpendicular to ground, compare the two-dimentional scaling board of plane in the past, it is accurate that three-dimensional scaling board is chosen special title point position, significantly reduced the distortion at fisheye camera imaging edge, imaging clearly, and error amount is low, and Image Mosaics precision significantly improves.
2, adopt virtual measurement technique, existing concept is applied to and measured in calculating, while demarcating a certain parameter, other parameters remain unchanged, and have promoted accuracy and the counting yield of calibrating parameters.
Embodiment
Below in conjunction with accompanying drawing, the present invention is done further and explained.
As shown in Figure 1, traditional vehicle-mounted viewing system is demarcated and is adopted two-dimentional scaling method, lays on the ground plane reference cloth.What use due to vehicle-mounted viewing system is the wide-angle camera of 170 degree left and right, and panorama picture of fisheye lens distortion is serious, and therefore this method makes to demarcate the very little and edge fog of cloth size of imaging in camera.
Calibration mark thing provided by the invention is placed perpendicular to ground, and the present embodiment adopts the scaling board of red color lump and blue color lump, and edge clear also can adopt other common scaling boards on the market.Intersect the demarcation cloth that in the past lay in ground, perpendicular to the unique point on the scaling board on ground, be positioned at same plane, reduced the camera distortion error that collection causes to unique point.Greatly increased the accuracy of proofreading and correct.The invention is characterized in, be not scaling board, can get a little the arbitrfary point on calibration mark thing.For example mark can be a triangle cone, and four summits, as unique point, are not limited only to selected characteristic point in the plane, can on three-dimensional object, get unique point.
The present embodiment provides a kind of vehicle-mounted viewing system camera parameter calibration method: it is characterized in that, when vehicle-mounted viewing system camera inner parameter and calibrating external parameters, adopt three-dimensional symbol mark to demarcate.
Three-dimensional symbol mark is positioned on three-dimensional correction mark, and the three-dimensional scaling coordinate system coordinate of this sign mark is known.
In viewing system, the FOV (Field of View) of each camera can clearly be seen 3 above described three-dimensional symbol marks.
Comprise vehicle-mounted viewing system and calibration mark thing; Described vehicle-mounted viewing system comprises 2-4 platform camera, and camera is placed in middle part all around, vehicle top;
The vehicle-mounted viewing system of this example comprises 4 cameras, and the camera of every camera is respectively the front camera (CF) that is positioned at headstock center position, the right camera (CR) that is positioned at the rear camera (CB) of tailstock center position, the left camera (CL) of vehicle left side side mirror position and side mirror position, vehicle right side;
Wherein, the FOV (Field of View) of front camera (CF) is VFF, and the FOV (Field of View) of rear camera (CB) is VFB, and the FOV (Field of View) of left camera (CL) is VFL, and the FOV (Field of View) of right camera (CR) is VFR; Set FOV (Field of View) overlapping region, VFF and VFL overlapping region are that VF-LF, VFF and VFR overlapping region are that VF-RF, VFL and VFB overlapping region are that VF-LB, VFR and VFB overlapping region are VF-RB;
Described camera is fish-eye camera.Select fish eye lens, can guarantee more wide FOV (Field of View).Existing vehicle-mounted its inner parameter of wide-angle camera on market is comprised: horizontal focal distance f x, longitudinal focal distance f y, the non-square degree alpha of sensitive chip pixel, the position cx of camera lens optical axis on sensitive chip,, cy.Radial distortion and tangential distortion k1, k2, k3, k4 and k5.For fish eye lens, according to the camera model of selecting, also have other model parameter to add, for fish eye lens camera, existed several universal models to select now,
This example is used the fisheye camera mathematical model in appeal background technology Literature 3.Adopt general mathematical model to avoid the dependence for specific mathematical model to the camera of particular vendors particular batch, thereby can when coding, not need the camera of certain model to write separately calibrating procedure.
This forwarding method comprises the following steps:
Step 1) put calibration mark thing: as shown in Figure 2, put calibration mark thing: demarcate on the ground a region that can hold vehicle, vehicle to be corrected is parked in this rectangular area; This region can be the shapes such as rectangle, ellipse, specifically do not limit, and for the ease of mark, preferred rectangle.Described three-dimensional correction mark comprises some unique points that can identify from image and that three-dimensional coordinate is known, such as angle point, and the unique points such as round dot;
The length of this rectangular area and width are greater than respectively length and the width of car body; Described calibration mark thing is scaling board, monochromatic color lump, and as red or blue, convenient demarcation and calculating.Scaling board has four groups, every group of four drift angle outsides that lay respectively at rectangular area; Every group is two, and the side of these two scaling boards is connected, and two scaling boards form " L " type; Two demarcation faces lay respectively on the long and wide extended line in rectangular area, and the demarcation face of scaling board is perpendicular to ground; The scaling board here, also can select the laying method of accompanying drawing 5, two scaling boards of same group is demarcated to faces and vertically place, be spaced a distance, respectively with the length of rectangular area with wide vertical.It is pointed out that this scaling board of placing perpendicular to ground not necessarily contacts with ground, also can adopt the mode of suspension to place.
Step 2) park vehicle to be corrected:
Sail vehicle to be corrected into step 1) in described rectangular area, adjust vehicle location; Carry out artificial or carry out automated graphics data acquisition by external trigger signal.When four FOV (Field of View) overlapping regions cover respectively one group of calibration mark thing, vehicle is come to a complete stop; In traditional 2DC method, vehicle must be parked in formulates position, and in this method, vehicle is as long as guarantee that each camera can absorb specified sign object image, without the position of special constraint vehicle parking, greatly improved convenience and the efficiency of operation.
Step 3) every camera calibration:
3.1) obtain calibration mark thing unique point: obtain calibration mark thing unique point: by feature point extraction way extract minutiae from each image, and mate with known unique point three-dimensional coordinate; Four summits that this example is set every scaling board are unique point; The scaling board that wherein in every camera FOV (Field of View), the relative two boards of demarcation face is every camera, take the calibration mark thing unique point that eight summits of these two scaling boards are every camera;
3.2) take step 3.1) two dimensional image coordinate and the corresponding known three-dimensional coordinate of the unique point obtained be measurement data, utilizes the inside and outside parameter of method of value solving calibration for cameras; By parameter grouping, by demarcating; While demarcating wherein one group of parameter, other parameters remain unchanged;
3.3) utilize step 3.2) the camera inside and outside parameter that obtains utilizes projective transformation technology to the mark unique point in three dimensions, projected on original image, calculates projection coordinate and the residual error of measuring coordinate;
3.4) default cycle index and residual error threshold value, judge whether to exceed cycle index threshold value or determining step 3.3) whether the residual error that obtains lower than predetermined threshold value;
If do not exceeded cycle index threshold value and residual error, be not less than predetermined threshold value, circulation step 3.1) to step 3.3);
As exceed cycle index threshold value or residual error lower than predetermined threshold value, judge whether residual error is greater than predetermined threshold value; As residual error is greater than predetermined threshold value, system provides the information of again demarcating, repeating step 3.1) to step 3.3); As residual threshold value is not more than predetermined threshold value, demarcate and finish, store and demarcate achievement.
This sentences calibration for cameras optimization external parameter is example, respectively the parameters such as camera external parameter and optical axis center point, inner parameter, distortion are calculated, as shown in the block diagram of Fig. 4, optimization camera external parameter pose: described external parameter comprises 3 displacement parameters and 3 rotation parameters; Displacement parameter Xgc,, Ygc and Zgc; Rotation parameter is respectively rxgc, rygc and rzgc; Demarcation for every camera, measure 8 unique points of two scaling boards, these unique points are 2-D data, i.e. 16 data, and the external parameter of required demarcation has 3 rotation parameters of 3 displacement parameters totally 6 parameters, the inner parameter of required demarcation also has 10 left and right according to above-mentioned to the elaboration of camera inner parameter, and the parameter that therefore altogether will demarcate has 16 left and right.For measuring dimension, be equal to or less than the situation of demarcating dimension, be easy to occur by the coupling between calibrating parameters (coupling) and over-fitting (over-fitting).
Optimization camera external parameter pose and optical axis center point cx, cy: by virtual measurement technique, measure three virtual measured quantity fy, cx and cy, its measured value is respectively fx, ImageWidth/2 and ImageHeight/2;
Wherein, fx is that the horizontal focal length of camera, fy are that the longitudinal focal length of camera, cx and cy are the position of camera lens optical axis on sensitive chip, the horizontal pixel number that ImageWidth is camera image, longitudinal pixel count that ImageHeight is camera image;
Here apply to priori, the horizontal focal distance f x of camera approaches with longitudinal focal distance f y numerical value, the position cx of camera lens optical axis on sensitive chip, and cy should be near the center of image.The present invention sees through just virtual measurement technique above-mentioned priori is joined in calibration process.
Optimization camera external parameter pose and the inner parameter except optical axis center point and distortion parameter:
Optimization camera external parameter pose distort k1, k2, k3, k4 and k5: wherein, k1, k2, k3, k4, k5 are radial distortion and tangential distortion
Above-mentioned parameter searching process can be used multiple existing algorithm to realize, as EKF, and Levenberg-Marquit algorithm etc.This sentences optimization camera external parameter pose is example explanation searching process, all the other processes are similar with it, individual in this step, optimizing object is (Xgc, Ygc, Zgc, rxgc, rygc, rzgc), rational initial value (Xgc0 of given optimizing object, Ygc0, Zgc0, rxgc0, rygc0, rzgc0), can be in the hope of optimizing object each unique point coordinate position on image on scaling board when this currency according to the mathematical description of geometrical optics projection process, and on scaling board the coordinate of each unique point with respect to optimizing object the rate of change at this currency, it is Jacobi matrix, to on real image, demarcate the position of plate features point and the coordinate position of above-mentioned calculating gained unique point on image do poor, be called residual error, Kalman filtering or Levenberg-Marquit algorithm by the residual sum Jacobi matrix numerical value that progressively iteration is upgraded optimizing object until residual error is less than given threshold value or iteration is greater than a certain given threshold value herein.The numerical value of the optimizing object obtaining after iteration renewal process finishes be the result of optimization procedure.
The present invention does in batches and demarcates all parameters, but calibrating parameters grouping is demarcated every group of parameter by entering, and when demarcating a certain group of parameter, other parameters keep original value constant.
As shown in Figure 6, the scaling board of take perpendicular to ground is example, the calibration range that 3D demarcates projects to the equivalent plane in plane, the scope of demarcating is demarcated much larger than the 2D of plane, as shown in Figure 1,2D demarcates the area of plane can only demarcate a plate, and the area of the dash area of onesize scaling board shown in can calibration maps 6 in space is demarcated is obviously better than 2D and demarcates.
The above is only the preferred embodiment of the present invention; it should be pointed out that for those skilled in the art, under the premise without departing from the principles of the invention; can also make some improvements and modifications, these improvements and modifications also should be considered as protection scope of the present invention.