CN104048662A - Robot vision homing method based on simplified road sign - Google Patents

Abstract

The invention discloses a robot vision homing method based on a simplified road sign. According to the robot vision homing method, a reflecting mirror and imaging equipment below the reflecting mirror are used for generating a panoramic image; the optical axis of the imaging equipment points to the reflecting mirror; the major and minor axes of the reflecting mirror are a and b; the distance from the intersection point of horizontal incident light and the reflecting mirror to the focal point of the reflecting mirror is r; the focal distance of the imaging equipment is f; a horizon circle is located on the panoramic image and takes the center of the horizon circle as a circle center; the horizon circle is expanded to form a final horizon ring region; SIFT characteristics are extracted in the final horizon ring region to be used as road sign points; the road sign points are used for calculating a homing vector of a robot. The robot vision homing method based on the simplified road sign has the advantages that the quantity of the road sign points is reduced, the correspondence of the road sign points is improved, and the homing success rate and the homing precision of an algorithm are improved.

Description

A kind of robot vision based on simplifying road sign method of going home

Technical field

The invention belongs to a kind of robot vision method of going home, particularly use and to simplify road sign, a kind of robot vision based on simplifying road sign method of going home.

Background technology

Intelligent mobile robot is that a class is passed through sensor senses surrounding environment and the state of self, and complex environment is understood and judged, carries out on this basis decision-making and planning, realizes object-oriented movement, thereby completes the robot of certain task.In the research of going home in vision, obtaining of visual information is emphasis.Because panoramic picture has the larger visual field, can obtain the environmental information that horizontal direction 360 is spent, therefore most of vision homing algorithms adopt panoramic picture.

In all kinds of vision homing algorithms, ALV algorithm has that model is simple, the better performances of going home and the advantage such as required storage space is little, and therefore this algorithm has consequence in the vision field of going home.In actual applications, ALV algorithm usually needs the local feature of environment for use image as natural landmark, and whole unique points of image are all used as road sign point, and this road sign point quantity that algorithm need be processed is more, affects algorithm counting yield.Meanwhile, in this case, the correspondence of road sign also cannot ensure, causes the precision of homing algorithm poor.Therefore study the problem of improving road sign correspondence and simplifying road sign quantity, go home precision and the counting yield of raising algorithm that can be larger.

Summary of the invention

The object of this invention is to provide and there is height a kind of robot vision based on simplifying road sign of precision method of going home of going home.

The present invention is achieved by the following technical solutions:

Robot vision based on the simplifying road sign method of going home,

Catoptron and be positioned at its below imaging device for generating panorama image, the optical axis directional mirror of imaging device, catoptron major and minor axis is respectively a, b, the intersection point of glancing incidence light and catoptron is r apart from the focal length of catoptron, the focal length of imaging device is f, horizon circle is positioned on panoramic picture, and taking the center of panoramic picture as the center of circle, horizon radius of a circle R is:

$R=\frac{r*f}{d}$

Catoptron equation is:

x ²/a ²-(y+c) ²/b ²＝-1

Wherein, $c=\sqrt{a^2+b^2},r=a*\sqrt{\frac{c^2}{b^2}-1},d=2*\sqrt{a^2+b^2};$

Horizon circle is expanded, formed final horizon ring territory;

In final horizon ring territory, extract SIFT feature as road sign point, utilize road sign point to calculate the vector of going home of robot.

A kind of robot vision method of going home based on simplifying road sign of the present invention can also comprise:

1, the method that forms final horizon ring territory is:

A, taking the center of concrete scene as the center of circle, set up rectangular coordinate system, the image of taking a concrete scene at the home position of coordinate system is as target location image, an image of the each shooting of boundary of the concrete scene in 4 quadrants of coordinate system is as current location image, each current location image and an image pair of target location image composition

B, in each image pair, in circle and outside circle, expand the width of horizon circle simultaneously, form horizon ring territory, determine the optimum width in horizon ring territory according to the variation tendency of road sign corresponding rate in horizon ring territory,

The maximal value of c, the optimum width of 4 images of selection to corresponding respectively horizon ring territory, as the final width in horizon ring territory, forms final horizon ring territory.

2, in final horizon ring territory, obtained n road sign point L by extracting SIFT feature ₁, L ₂... L _n, the position of road sign point in image is X _i=(x, y), i=1,2...n, the position of robot is X, robot at the unit road sign vector of X position is:

$\stackrel{\→}{{\mathrm{LV}}_i\left(X\right)}=\frac{X_i-X}{\left|\right|X_i-X\left|\right|}=({\cos \mathrm{\θ}}_i,{\sin \mathrm{\θ}}_i),$

The average road sign vector of X position is:

$\stackrel{\→}{\mathrm{ALV}\left(X\right)}=\frac{1}{n}\underset{i=1}{\overset{n}{\mathrm{\Σ}}}\stackrel{\→}{{\mathrm{LV}}_i\left(X\right)}$

The vector of going home that utilizes road sign point to calculate robot is:

$\stackrel{\→}{\mathrm{\β}\left(H\right)}=\stackrel{\→}{\mathrm{ALV}\left(C\right)}-\stackrel{\→}{\mathrm{ALV}\left(H\right)},$

Wherein, for the average road sign vector of current location, for the average road sign vector of target location.

Beneficial effect of the present invention is:

The advantage of invention is to have reduced road sign point quantity, has improved road sign point correspondence, has improved go home success ratio and the precision of going home of algorithm.For effect of the present invention is described, the present invention adopts the panoramic picture database that University Bielefeld provides to analyze, and this database is widely used in the go home test of method of vision.In database, choose original, a chairs and day3 scene, go home success ratio correlation data as shown in Fig. 1 (a)～Fig. 1 (c), Fig. 2, Fig. 3 for 15 target locations.

Brief description of the drawings

Fig. 1 (a) is the original scene success ratio statistical form of going home, and Fig. 1 (b) is the chairs scene success ratio statistical form of going home, and Fig. 1 (c) is the day scene success ratio statistical form of going home.

Fig. 2 is the vector plot of going home that ALV algorithm calculates.

Fig. 3 is the vector plot of going home that the inventive method calculates.

Fig. 4 is the schematic diagram that calculates Horizon circle imaging radius.

Fig. 5 is imaging schematic diagram.

Fig. 6 is that the coordinate of the sample space of test pattern database is described.

Fig. 7 (a)～Fig. 7 (d) is the corresponding variation relation that road sign corresponding rate and horizon encircle field width degree, Fig. 7 (a) is that picture position is C (1,3), H (5,9) result, Fig. 7 (b) is that picture position is C (0,16), H (5,9) result, Fig. 7 (c) is that picture position is C (8,2), H (5,9) result, Fig. 7 (d) is that picture position is C (9,13), the result of H (5,9).

Fig. 8 is average road sign vector ALV model.

Embodiment

Below in conjunction with accompanying drawing, the present invention is described in further details.

A kind of ALV algorithm of simplifying road sign based on refractive and reflective panorama image has been proposed.This invention, according to the feature of the refraction-reflection panoramic imagery based on hyperbolic mirror, has proposed the concept in horizon ring territory, utilizes the more stable characteristic of unique point in ring territory, and the SIFT feature of extracting in ring territory calculates as road sign and in conjunction with ALV model the direction of going home.The homing algorithm that the present invention proposes has ensured the correspondence of road sign preferably, effectively raises and goes home precision and reduced calculated amount.Through actual scene image authentication, algorithm performance is reliable and stable.Comprise following step:

Step 1, determine horizon circle by the panoramic picture of concrete scene and the parameter of omnidirectional imaging system.As shown in Figure 1: establish catoptron major and minor axis and be respectively a, b, the intersection point of glancing incidence light and catoptron is r apart from the focal length of catoptron, the focal length that f is camera, R is horizon radius of a circle.As shown in Figure 4.

Catoptron equation is:

X ²/ a ²-(y+c) ²/ b ²=-1, wherein

Calculate known:

$r=a*\sqrt{\frac{c^2}{b^2}-1};d=2*\sqrt{a^2+b^2};$

Known according to the character of similar triangles:

$\frac{R}{f}=\frac{r}{d}\⇒R=\frac{r*f}{d}$

After calculating R, taking picture centre as the center of circle, the circle that R is radius is horizon circle.

Step 2, determine horizon circle by step 1 after, taking the center of actual scene as the center of circle, set up rectangular coordinate system, horizontal direction does not need concrete restriction.The image of taking an actual scene at the home position of coordinate system, as target location image, then approaches the each image of taking an actual scene in scene boundary place as current location image in 4 quadrants of this coordinate system, takes particular location arbitrarily.By 4 current location images respectively with target location image composition diagram picture pair, determine horizon circle in each image pair, and taking horizon circle as benchmark, taking pixel as unit, in circle and outside circle, both direction is expanded the width formation annular region of horizon circle simultaneously, is called horizon ring territory in the present invention.Extract the SIFT feature of image as road sign point, definition road sign corresponding rate is the coupling road sign point quantity in horizon ring territory and whole ratio of road sign point quantity that mates between two width images.Corresponding to 4 images pair, make respectively the right horizon ring field width degree of each image and the graph of relation of road sign corresponding rate, taking slope of a curve, as 0 o'clock, corresponding ring field width degree, as optimum width, was got the final width of 4 maximal values in optimum width as horizon ring territory for the first time.

Step 3, in step 2, in definite horizon ring territory, extract SIFT feature as road sign point, and these road sign points are applied to and in ALV model, calculate the final vector of going home.

Below the principle of the specific implementation process of main this part in the present invention is introduced respectively:

Determining of 1.Horizon ring territory and optimum width

In refraction-reflection omnidirectional imaging system, curved mirror is used to form the image of surrounding environment scene, it be positioned at its below and optical axis and point to the collection that completes panoramic picture together with its video camera.In the ideal case, in the panoramic picture not launching, have a circle taking picture centre as the center of circle, the road sign spot projection being positioned in same level with curved mirror must be positioned at this circle above, as shown in Figure 5.The space of supposing robot is plane, and, along with the motion of robot, these road sign points can move or be blocked on this circle, but can not move to outside circle.In the present invention, claim this circle for horizon circle.

Although the road sign point on horizon circle is more stable, horizon circle only has the width of a pixel, may only extract little road sign point on circle, extracts less than road sign point even at all.Owing to there being the problems referred to above, the present invention expands the width of horizon circle, and we claim that the annular region after expansion is horizon ring territory.In order to determine the optimum width in horizon ring territory, the present invention adopts the panoramic picture database that University Bielefeld provides to analyze, and this database is widely used in the test of vision homing algorithm.According to image data base sampling characteristics, adopt the sampling location of the system of discrete coordinates shown in Fig. 6 presentation video herein, for example in Fig. 6, in black round dot representative image sampling scene, coordinate is the position of (5,9).

Before testing, following two explanations of given first:

(1) for the position coordinates in sampling scene, target location adds H and represents before coordinate, and current location adds C and represents before coordinate.For example, H (5,9) represents that the coordinate of target location is (5,9).

(2) in the present invention, road sign corresponding rate is defined as the coupling road sign point quantity in horizon ring territory and whole ratio of road sign point quantity that mates between two width images.

We choose 5 images in image library, corresponding scene location coordinate is respectively (1,3), (0,16), (5,9), (8,2) and (9,13), in order to get rid of to greatest extent the impact of contingency, select (5,9) be target location, all the other 4 positions lay respectively in 4 quadrants of (5,9) as current location.By experiment, we have obtained the relation of road sign corresponding rate and horizon ring field width degree, as shown in Fig. 7 (a)～Fig. 7 (d).In Fig. 7 (a)～Fig. 7 (d), can find out, corresponding to the image of different current locations, just start the increase gradually along with horizon ring field width degree, road sign corresponding rate also increases gradually, in the time that ring field width degree approaches 30 pixels, road sign corresponding rate is steady gradually, thereafter along with encircling the increase of field width degree, road sign corresponding rate changes very little, even substantially constant.Therefore ring field width degree corresponding when, road sign corresponding rate is steady is optimum width.

2. the vectorial calculating of going home

Suppose to have extracted n road sign point L in horizon ring territory ₁, L ₂... L _n, in the robot visual field, pointed to the vector of unit length of corresponding road sign by robot location be defined as road sign vector, as shown in Figure 8.

If n road sign position in image is X _i=(x, y), i=1,2...n, the position of robot is X, unit road sign vector is

$\stackrel{\→}{{\mathrm{LV}}_i\left(X\right)}=\frac{X_i-X}{\left|\right|X_i-X\left|\right|}=({\cos \mathrm{\θ}}_i,{\sin \mathrm{\θ}}_i),$

Average road sign vector in X position is

$\stackrel{\→}{\mathrm{ALV}\left(X\right)}=\frac{1}{n}\underset{i=1}{\overset{n}{\mathrm{\Σ}}}\stackrel{\→}{{\mathrm{LV}}_i\left(X\right)}$

Being located at the average road sign vector that current location and target location calculate is respectively with the vector of going home by with subtraction of vector obtain

$\stackrel{\→}{\mathrm{\β}\left(H\right)}=\stackrel{\→}{\mathrm{ALV}\left(C\right)}-\stackrel{\→}{\mathrm{ALV}\left(H\right)}$

Calculate final going home after vector, robot along direction move and can get back to target location.

Claims (3)

1. the method for going home of the robot vision based on simplifying road sign, is characterized in that:

Catoptron and be positioned at its below imaging device for generating panorama image, the optical axis directional mirror of imaging device, catoptron major and minor axis is respectively a, b, the intersection point of glancing incidence light and catoptron is r apart from the focal length of catoptron, the focal length of imaging device is f, horizon circle is positioned on panoramic picture, and taking the center of panoramic picture as the center of circle, horizon radius of a circle R is:

$R=\frac{r*f}{d}$

Catoptron equation is:

x ²/a ²-(y+c) ²/b ²＝-1

Wherein, $c=\sqrt{a^2+b^2},r=a*\sqrt{\frac{c^2}{b^2}-1},d=2*\sqrt{a^2+b^2};$

Horizon circle is expanded, formed final horizon ring territory;

In final horizon ring territory, extract SIFT feature as road sign point, utilize road sign point to calculate the vector of going home of robot.

2. a kind of robot vision based on simplifying road sign according to claim 1 method of going home, is characterized in that: the method in the final horizon ring of described formation territory is:

A, taking the center of concrete scene as the center of circle, set up rectangular coordinate system, the image of taking a concrete scene at the home position of coordinate system is as target location image, an image of the each shooting of boundary of the concrete scene in 4 quadrants of coordinate system is as current location image, each current location image and an image pair of target location image composition

B, in each image pair, in circle and outside circle, expand the width of horizon circle simultaneously, form horizon ring territory, determine the optimum width in horizon ring territory according to the variation tendency of road sign corresponding rate in horizon ring territory,

The maximal value of c, the optimum width of 4 images of selection to corresponding respectively horizon ring territory, as the final width in horizon ring territory, forms final horizon ring territory.

3. a kind of robot vision based on simplifying road sign according to claim 2 method of going home, is characterized in that: in final horizon ring territory, obtained n road sign point L by extracting SIFT feature ₁, L ₂... L _n, the position of road sign point in image is X _i=(x, y), i=1,2...n, the position of robot is X, robot at the unit road sign vector of X position is:

$\stackrel{\→}{{\mathrm{LV}}_i\left(X\right)}=\frac{X_i-X}{\left|\right|X_i-X\left|\right|}=({\cos \mathrm{\θ}}_i,{\sin \mathrm{\θ}}_i),$

The average road sign vector of X position is:

$\stackrel{\→}{\mathrm{ALV}\left(X\right)}=\frac{1}{n}\underset{i=1}{\overset{n}{\mathrm{\Σ}}}\stackrel{\→}{{\mathrm{LV}}_i\left(X\right)}$

The vector of going home that utilizes road sign point to calculate robot is:

$\stackrel{\→}{\mathrm{\β}\left(H\right)}=\stackrel{\→}{\mathrm{ALV}\left(C\right)}-\stackrel{\→}{\mathrm{ALV}\left(H\right)},$

Wherein, for the average road sign vector of current location, for the average road sign vector of target location.