CN104792312A - Indoor automatic transport vehicle positioning system with three fixed balls as visual marker - Google Patents

Abstract

The invention discloses an indoor automatic transport vehicle positioning system with three fixed balls as a visual marker. The system is characterized in that three luminescence balls with specific size proportions and specific positions are combined as a marker, and are fixed on an automatic transport vehicle; a pinhole camera is fixed on an indoor ceiling, and downward shoots the marker; and an embedded system connected to the camera is used to carry out marker extraction and analysis of every frame of images in order to complete detection of the pose of the automatic transport vehicle, and communication and control of the automatic transport vehicle are completed through a wireless module according to the above positioning result. The positioning system has the advantages of simple calculation, simultaneous realization of position, direction and inclination angle positioning, good stability, high precision, and completion of positioning in complex environment.

Description

With the indoor motor dolly positioning system that spacing three ball is visual beacon thing

Technical field

The present invention relates to a kind of indoor positioning technologies field, particularly a kind of indoor motor dolly field of locating technology based on machine vision.

Background technology

Owing to there is more restrictive condition in indoor environment, cause gps system can not play the advantage of its accurate location.Therefore find an appropriate technology and carry out the focus that indoor positioning (IPS) also becomes current research and development very soon.Indoor positioning technologies conventional at present comprises AGPS technology, ultrasonic technology, REID and wireless local area network technology etc., but the positioning precision of these technology is lower, not only be subject to multipath effect and environmental turbulence and affect calculating, and need in a large amount of hardware facility auxiliary positioning of indoor layout, be unfavorable for safeguarding.There is the object space of some view-based access control model and the detection method in direction in recent years, if document " DeRen Li; Yong Liu; XiuXiao Yuan; Image-based self-position and orientation method formoving platform [J] .Science China Information Sciences; 2013; 56 (4): 4649-4662. " its main advantage to avoid the error of above-mentioned technology because of transmission generation, higher positioning precision is provided, and the detection that hardware cost just can carry out object direction and inclination angle can not be increased.

For the locator meams of machine vision, mainly contain following three kinds of schemes at present: monocular vision location, binocular visual positioning and omni-directional visual location.Based on the positioning system of binocular vision, mainly obtain the parallax information of scene image according to dual camera and carry out position judgment, as document " Bai Ming; solemn; Wang Wei. the research and advances [J] of Binocular Stereo Matching Algorithm. control and decision-making, 2008,23 (7): 721-729. "; Omni-directional visual positioning system, omnidirectional camera is utilized to obtain the image comprising 360 ° of scene informations, then comprehensive figure is launched into Cylindrical panoramic image, the location of last realize target thing, as document " Zhou Minghui. study [D] based on the cylinder unwrapping of omni-directional visual and Robust Real-time Moving Object Tracking. Shanghai: Shanghai Communications University, 2013 ".Contrast above two schemes, based on the positioning system of monocular vision without the need to solving pattern distortion in the pixel matching of dual image in binocular vision system and fully-directional visual system, pixel repeats and the error problem brought such as to lose, and have the advantages that cost is low, volume is little, real-time is high, be therefore used more and more in recent years.Monocular vision distance-finding method wherein based on geometric projection is widely used in mobile robot's range finding, the automatic obstacle of automobile etc.But when practical application, robot or walking about of automobile easily cause the shake of camera, and this will have a strong impact on location algorithm, cause larger error.

Usual motor dolly (AGV) itself is not suitable for as localizing objects, because the AGV outward appearance of different application is not quite similar, need to change recognizer for different AGV, and AGV outward appearance does not have unchangeability under different views, need to increase correction algorithm, this not only can bring rectification error, and can increase code overhead.

Summary of the invention

The object of the invention is the error problem in order to avoid above-mentioned existing indoor positioning technologies exists, there is provided a kind of and adopt fixing single camera mode and using three ball marks of particular combination as the indoor motor dolly positioning system of localizing objects, realize indoor motor dolly quick, accurate, locate reliably.

Technical scheme that the present invention adopts is as follows:

Design a kind of three ball marks by particular combination as localizing objects, the detection at the indoor automatic transportation truck position of indirect realization, direction and inclination angle, the large ball that this mark is the little radius of a ball 2 times by the identical bead of two radiuses and radius combines, and two bead centre ofs sphere and large ball centre of sphere distance are all 4 times of the large radius of a ball, the three ball centre ofs sphere form one with the distance of the bead centre of sphere and the large ball centre of sphere for waist is long, drift angle is the isosceles triangle of 120 °, three balls all can be respectively charged into the LED of three kinds of different colours, mix into the mark of 27 kinds of different codings.In flooring, the large ball centre of sphere is mainly as the position judgment of localizing objects, isosceles triangle in the orientation of flooring as the judgement in localizing objects direction, the centre of sphere of three games played with small balls form angle that plane and indoor plane produce and judge as the inclination angle of localizing objects.

The LED of described three kinds of different colours is the LED of red, green, blue three kinds of colors.

Described isosceles triangle orientation refers to the large ball centre of sphere for initial point, the direction that the ray formed with base mid point points in flooring.

(1) in indoor ceiling, fixedly mount the video camera of optical axis perpendicular to flooring, flooring image is obtained by this video camera, processor carries out pre-service, the identification of mark and the calculating of pose to obtained image, and completes navigation to motor dolly and control by wireless communication module.

Preferably, described processor is DSP.

The concrete processing procedure of described Image semantic classification is as follows: first carry out color segmentation to RGB image, then carries out morphological erosion and expansive working to the image after color segmentation, finally carries out edge extraction operation to image.

The concrete processing procedure of identification of described mark is as follows: first sort out and record all edge contour pixels, then the profile meeting area requirements and circularity requirement is screened, contour images after screening is carried out to the ellipse fitting of least square method, finally utilize the distance between three balls, dimension proportionate relationship and method for mode matching to identify mark.

The concrete processing procedure of calculating of described pose is as follows: first calculate three balls in position, flooring by the position of the three ball centre ofs sphere in image according to the geometric projection relation of pinhole imaging system, then draw direction and the inclination angle of object further according to the result of calculation of three balls.

Preferably, described wireless communication module is the radio transmitting and receiving chip being operated in 2.4G ~ 2.5GHz.

Compared with prior art, the present invention has the following advantages:

(1) carry out location, position to single sphere, it all has a unchangeability under visual angle, and arbitrary orientation does not affect the stability of identification, avoids the code overhead increasing correction algorithm and bring, improves recognition speed.

(2), when detecting motor dolly direction and inclination angle, only can need be detected by three ball position relations, without the need to processing motor dolly profile, this mode realizes succinctly, and error is little, real-time is good.

(3) according to three balls can photism, specific distance and dimension proportionate relationship, not only increase the reliability of system to Marker Identity, antijamming capability is strong, the location under complex environment can be completed, and may be combined with into the mark of 27 kinds of numberings, navigation and control while being applicable to multiple stage motor dolly.

(4) use of mark allows system be not limited to the motor dolly of certain shape, only mark need be arranged on motor dolly and just can locate, and improves system universality.

(5) adopt fixed cameras mode, walk about cause camera shake the error problem caused without the need to solving robot or automobile, effectively improve the positioning precision of indoor.

Accompanying drawing explanation

Fig. 1 is three ball mark organigrams in the present invention;

Fig. 2 is the marker position positioning principle illustraton of model in the present invention;

Fig. 3 is the mark angle detecting schematic diagram in the present invention;

The y direction that Fig. 4 (a) ~ (b) is respectively bead Q in mark tilts and two bead x direction inclination schematic diagram;

Fig. 5 is the system composition schematic diagram in the present invention;

Fig. 6 is that processor in the present invention is to the process of every two field picture and recognizer process flow diagram;

Embodiment

Below in conjunction with accompanying drawing, enforcement of the present invention is described in further detail

(1) a kind of three ball marks by particular combination are designed as localizing objects, the positioning function of the indoor motor dolly of indirect realization, this mark structure as shown in Figure 1, be R1:R2:R3=2:1:1 by three bead radius scale, two bead centre ofs sphere and large ball centre of sphere distance L0 and large radius of a ball R1 ratio are L0:R1=4:1, angle ∠ QPQ '=120 ° between two bead centre ofs sphere and the large ball centre of sphere, three balls all can be respectively charged into the LED of three kinds of different colours, mix into the mark of 27 kinds of various combinations.

When positioning the position detection of target, only need find out large ball centre of sphere position in the picture, then obtain the position in flooring by geometric projection relation.As shown in Figure 2, this camera imaging model is reduced to national forest park in Xiaokeng, in indoor coordinate system, using the vertical projection point O of camera in flooring as coordinate origin, right along camera direction is defined as X-direction, and camera dead ahead is defined as Y direction.In image coordinate system, H and W is respectively the height and width of image, and for convenience of calculating, define the initial point that the mid point G ' of image rectangle is image coordinate system, y-axis is that initial point is perpendicular to direction upwards, image coboundary.P (x, y) is the indoor coordinate of impact point, and p (u, v) is corresponding image coordinate.Then impact point formula below the coordinate y of Y-axis has is determined:

$y=(h-\mathrm{\Δh})\tan ({\tan }^{-1}\left(\frac{y_1}{h}\right)+{\tan }^{-1}\left(\frac{2v\×\tan \mathrm{\θ}}{H}\right))$

Wherein h is the height on camera distance ground, and y1 is the distance that camera optical axis and ground intersection point G to O put, and the vertical field of view angle that the height that Δ h is mark centroid distance ground, 2 θ are video camera, H is the height of image.

Then impact point formula below the coordinate x of X-axis has is determined:

$x=\frac{y}{\sin ({\tan }^{-1}\left(\frac{y_1}{h}\right)+{\tan }^{-1}\left(\frac{2v\×\tan \mathrm{\θ}}{H}\right))}\×\frac{2u\×\tan {\mathrm{\θ}}^{\′}}{W}$

Wherein 2 θ ' the horizontal field of view angle that is video camera, W is the wide of image.

With reference to Fig. 3, when positioning the angle detecting of target, first obtain the coordinate P (x1 of three ball centre ofs sphere at indoor coordinate system by above-mentioned formula, y1), Q (x2, y2), Q ' (x3, y3), before travel direction resolves, need to revise these coordinates, this is because when three ball centre of sphere planes and indoor plane not parallel (namely there is inclination angle), at this moment P (x1, y1), Q (x2, y2), Q ' (x3, y3) triangle formed to be no longer drift angle the be isosceles triangle of 120 °, but known this error is because the relatively large ball of bead has vertical shift to cause, make to calculate x, the result of calculation of y is changed simultaneously, and known three balls form a standard length of side is the isosceles triangle of L0, therefore can correct the triangle that there is distortion according to these characteristics, its method: calculate respectively with 2 range formulas distance Lpq and distance Lpq ', first try to achieve the deviation delta L of Lpq and L0, keep v constant, u reduces or increases by 1, again the deviation delta L ' of Lpq and L0 is calculated to obtain, find out the variation tendency of u when the relative Δ L of Δ L ' is reduced, and keep this trend, and same, keep u constant, change the value of v, the variation tendency of v when the relative Δ L of Δ L ' is reduced, continues to reduce or increase by 1 to v, u, until the departure of Lpq and L0 is minimum according to this trend, then stop circulation, again calculate to obtain coordinate figure y2, x2 of Q according to new v, u value.In like manner Q ' coordinate is corrected and obtain new coordinate figure y3, x3.If the value of certain Lpq and Lpq ' is at the beginning all close or equal L0, then without the need to correcting, directly the direction of carrying out below is resolved.

With large ball sphere centre coordinate P for limit sets up polar coordinate system, wherein pole axis is parallel to CD.Q (x2, y2), Q ' (x3, y3) rectangular coordinate transfer polar coordinates to, try to achieve line segment mid point Qm, Qm is the deflection β of mark at the polar angle of polar coordinate system, thus obtains the direction of motor dolly.

With reference to Fig. 4, when positioning the inclination angle detection of target, the centre of sphere of Main Basis three games played with small balls form the angle that plane and indoor plane produce, this angle is be decomposed in the rectangular coordinate system of initial point on the tilt angle gamma y(of y direction to have a down dip at P) and x direction tilt angle gamma x(lateral tilting), still first obtain indoor three ball sphere centre coordinate value P (x1 after correcting, y1), Q (x2, y2), Q ' (x3, y3), the y direction inclination schematic diagram that Fig. 4 (a) is bead Q, the relative P of Q distance Lqp=|y2-y1| in the y-axis direction, α q is tried to achieve by formula α q=tan-1 ((h-Δ h)/y2), Lq by formula Lq=L0 × | sin (β-π/3) | determine, then Q is determined by following formula at y direction tilt angle gamma yq:

${\mathrm{\γ}}_{\mathrm{yq}}={\sin }^{-1}\left(\frac{L_{\mathrm{qp}}\×\sin (180-{\mathrm{\α}}_q)}{L_q}\right)-{\mathrm{\α}}_q$

Bead Q relatively large ball P vertical offset Δ hq is determined by formula Δ hq=Lq*sin γ yq.

In like manner Q ' is determined by following formula at inclination angle, y direction:

${\mathrm{\γ}}_{{\mathrm{yq}}^{\′}}={\sin }^{-1}\left(\frac{L_{q^{\′}p}\×\sin (180-{\mathrm{\α}}_{q^{\′}})}{L_{q^{\′}}}\right)-{\mathrm{\α}}_{q^{\′}}$

Wherein Lq ' p is the Q ' distance of P point in y-axis direction relatively, α q '=tan-1 ((h-Δ h)/y3), Lq '=L0*|sin (β+π/3) |.

Bead Q ' relatively large ball P vertical offset Δ hq ' is determined by formula Δ hq '=Lq ' * sin γ yq '.

Therefore inclination angle, mark y direction is determined by following formula:

be inclined upwardly when γ y>0 represents, γ y<0 represents downward-sloping.

Fig. 4 (b) is two bead x direction inclination schematic diagram, wherein determines,

Finally when γ x>0 represents that Q is higher than Q ', γ x<0 represents that Q is lower than Q '.

(2) Fig. 5 gives of the present invention with the schematic diagram of the spacing three ball indoor motor dolly positioning system that is visual beacon thing, and with reference to Fig. 5, this system comprises: be fixed on the pinhole camera 1 in indoor ceiling 5; For the embedded system 2 of image procossing and identification; Three ball composite marker things 3 of motor dolly are fixed on as localizing objects; For the wireless module 4 that embedded system communicates with motor dolly.System work process: the optical axis of video camera 1 is perpendicular to flooring 6, downward shooting flooring image, the embedded system 2 being connected to video camera processes obtained image and analyzes, the identification of complement mark thing 3 and the calculating of pose, and complete navigation to motor dolly and control by wireless communication module 4.

In the present invention, embedded system process device to the process of image and recognizer flow process as shown in Figure 6, specifically comprises as follows:

Step 1: in R, G, B component of each pixel, if meet one of them component when being all greater than another two component 150 gray scale differences, is set to 255 this pixel, if do not met, sets to 0, carries out color segmentation according to this mode to image.

Step 2: for reducing the impact of noise and the red block of some small sizes, morphological erosion and expansive working are carried out to the image after color segmentation, here adopt the circular configuration element of 5 × 5 as form core, this not only removes noise preferably, and the not shape of mark in effect diagram picture.

Step 3: adopt Canny method, contour edge extraction is carried out to image.

Step 4: for removing the impact of class ellipse or other shape edges features in background, first contour images usable floor area is retrained, for the image of 640*480 pixel, here minimum area threshold Smin=40 and maximum area threshold Smax=31415 is chosen, the profile meeting interval S ∈ [Smin, Smax] of area is screened.

Step 5: after area screening, then adopt the ratio C=P2/S between girth square and area to carry out circularity judgement, wherein P represents the girth of object, and when profile meets circle characteristic, the value of trying to achieve C will be the value close to 4 π.

Step 6: obtain the contour images after screening, the identification justified can be carried out, adopt least square method ellipse fitting method here, all pixels of each edge contour are substituted into objective function respectively:

$\begin{array}{c}f(A,B,C,D,E,F)\\ =\underset{i=1}{\overset{N}{\mathrm{\&Sigma;}}}{({\mathrm{Ax}}_i^2+B{x}_i{y}_1+{\mathrm{Cy}}_i^2+{\mathrm{Dx}}_i+{\mathrm{Ey}}_i+F)}^2\end{array}$

Oval parameters A, B, C, D, E, F is confirmed by the minimum value solving objective function.

Step 7: for improving system identification reliability further, can be R1:R2:R3=2:1:1 according to three bead radius scale, two bead centre ofs sphere and large ball centre of sphere distance L0 and large radius of a ball R1 ratio are this particular geometric relation of L0:R1=4:1, combine three contiguous circles successively in image after circle matching, and carry out a geometry judgement and screening to often organizing three circles.

Step 8: often organize three circle in each circle adopt Hu invariant moments matching to carry out region shape coupling respectively, the method has stronger translation, yardstick and rotational invariance, and matching result is determined by following formula:

$I(A,B)=\underset{i=1}{\overset{7}{\mathrm{\&Sigma;}}}\frac{|m^{A_i}-m^{B_i}|}{\left|m^{A_i}\right|}$

$m^{A_i}=\mathrm{sign}\left(h^{A_i}\right)\&CenterDot;\log \left(h^{A_i}\right),m^{B_i}=\mathrm{sign}\left(h^{B_i}\right)\&CenterDot;\log \left(h^{B_i}\right),$

Wherein it is the Hu square of A and B

Specific practice is as follows:

Suppose to there are N number of three circles meeting ratio condition, the each round center (u of three circles often organized in record, v) i={ (ui, vi), (ui+1, vi+1) }, i ∈ [0, N] and radius R i={ri, ri+1}, i ∈ [0, N], centered by each round center, long and wide is all that the rectangle of 2R arranges area-of-interest (ROI) at former RGB image, and the maximum passage of the RGB component taking out each ROI carries out binary conversion treatment, and record this ball color Ci, be black outside last and round, for the template of white carries out Hu match by moment in circle, that only has each circle in three circles to meet simulated condition is only final recognition result simultaneously, and record the center of circle p (u1 of now obtained large ball image, and color C1 v1), according to the center of circle q (u2 counterclockwise recording Q bead image, and color C2 v2), center of circle q ' (u3 of Q ' bead image, and color C3 v3).

Often kind of color two bits represent, here specify that 00 represents redness, 01 represents green, 10 represent blueness, therefore three ball color combination need to represent with 6 bits, such as C1C2C3 is respectively blue, green, red, then this combination binary coding is 100100, is numbered the indoor motor dolly of multi rack by which.

Step 9: the image coordinate p (u1 of the three ball centre ofs sphere, v1), q (u2, v2), q ' (u3, v3) the indoor coordinate P (x1 that namely (1) described method can obtain the three ball centre ofs sphere is substituted into, y1), Q (x2, y2), Q ' (x3, y3), and then obtain position P (x, y), the deflection β, up and down tilt angle gamma y, left and right tilt angle gamma x of motor dolly.

Above-described embodiments of the present invention, do not form limiting the scope of the present invention.Any amendment done within spiritual principles of the present invention, equivalent replacement and improvement etc., all should be included within claims of the present invention.

Claims (1)

1. with the indoor motor dolly positioning system that spacing three ball is visual beacon thing, to it is characterized in that by three ball marks of particular combination as localizing objects, indirectly realize the detection at indoor automatic transportation truck position, direction and inclination angle;

The large ball that three ball marks of described particular combination are the little radius of a ball 2 times by the identical bead of two radiuses and radius combines, and two bead centre ofs sphere and large ball centre of sphere distance are all 4 times of the large radius of a ball, the three ball centre ofs sphere form one with the distance of the bead centre of sphere and the large ball centre of sphere for waist is long, drift angle is the plane isosceles triangle of 120 °, this plane is parallel with motor dolly chassis plane, three balls all can be respectively charged into the LED of three kinds of different colours, mix into the mark of 27 kinds of different numberings;

Described large ball is used for the position judgment of mark in flooring;

Described isosceles triangle is used for the detection in mark direction, namely in isosceles triangle summit be the point of the large ball centre of sphere as initial point, the direction that the ray formed with base mid point points in flooring is mark direction;

Described plane is used for mark inclination angle detection, i.e. the centre of sphere of three games played with small balls form angle that plane and indoor plane the produce inclination angle as mark;

Described difference is numbered employing 00 and represents redness, and 01 represents green, and 10 represent blueness, uses 6 bits to represent the mark that different colours combines.