CN103983270A - Graphic sonar data processing method - Google Patents

Abstract

The invention discloses a graphic sonar data processing method which can be effectively applied to mobile robot localization, map building and route planning processes. The method comprises the following steps: performing noise reduction treatment on sonar data, extracting angular point information from the data, and expanding an angular point range based on an error ellipse and a largest error circle so as to build a local expansion map; mapping the local expansion map to a binary image space, establishing a local expansion map in the binary image space, and matching the current local expansion map with a historical local expansion map through a rotational invariance matching method, a two-dimensional scanning matching method and a scale invariant feature transform (SIFT) matching method. Compared with a traditional direct sonar data processing method, the novel sonar distance data processing method provided by the invention has the advantages that the information contained in the sonar data can be fully explored due to graphic sonar data processing, the sonar data processing mode is enriched, and the precision and robustness of sonar data processing are improved.

Description

A kind of image conversion disposal route of sonar data

Technical field

The present invention relates to a kind of disposal route of sonar range data, by range data being mapped to the mode of image space, realized the image conversion of sonar data and processed.Belong to Mobile Robotics Navigation field.

Background technology

Along with the progress of airmanship, mobile robot is widely used to assist the mankind to complete the tasks such as circumstances not known detection, and " Jade Hare " lunar rover of China is typical example.The inter-related task of surveying for completing circumstances not known, robot need possess the function of independent navigation.Generally, does independent navigation comprise three subproblems: 1. I am at that? which will 2. I go? which do 3. how I go? the corresponding robot of difference location, map building and path planning.The perception of robot self pose and environmental characteristic positional information is the prerequisite addressing the above problem.

At present, mobile robot mainly uses vision sensor, laser sensor and ultrasonic sensor etc. to obtain the positional information of self and environmental characteristic.Wherein, the abundant information that vision sensor obtains, but require robot to possess data-handling capacity fast, in addition, vision sensor disturbs responsive to light, block etc., limited its range of application.Laser sensor and ultrasonic sensor are range sensor, by the distance between robot measurement and environmental characteristic, provide robot navigation required information.Laser sensor response is fast, and the precision of information of obtaining is high, and still, laser sensor installation accuracy requires high, expensive.Relative, ultrasonic sensor is installed simple, and price is relatively low, can obtain the information that precision is relatively high, and therefore, ultrasonic sensor is still widely used.But because field angle is larger, the information that ultrasonic sensor obtains exists uncertain.Probability theory, fuzzy theory and gray system theory etc. are all used to express, process ultrasound information, and finally realize robot map building, location and path planning.

Taking a broad view of current sonar information by using disposal route, is generally directly to process the range data that sonar obtains, and by calculating the statistical information of range data, sets up the formalized description of environmental characteristic.Typical example is as the rasterizing describing method of environment and characteristics map creation method.Affected by field angle, the range information that sonar obtains is inevitably with error.The statistical information that current sonar range data processing method has utilized packet to contain to greatest extent, and in mobile robot autonomous navigation, obtained successful Application.But when error is larger, the accuracy meeting of statistical information is affected, meanwhile, existing method is also difficult to further from raw data, excavate useful information.For this reason, sonar range data are mapped to image space from metric space herein, the related art method of utilizing image to process realizes the processing of sonar range information.The method can effectively be applied in localization for Mobile Robot, map building and path planning process.

Summary of the invention

The object of the invention is the range data of sonar to map to image space, utilize image processing techniques to realize the processing of sonar range data, by the processing of image conversion, excavate to greatest extent the environmental information that sonar data comprises, the new approaches that one aspect of the present invention provides sonar data to process, on the other hand, by the image conversion matching technique of multiple sonar data, precision and robustness that sonar data is processed have been improved.

The invention provides a kind of image conversion disposal route of sonar data, by sonar range information is mapped to image space, the method for utilizing image to process, realizes the processing of sonar range information, mainly comprises the following steps:

Step 1, filter owing to measuring blind area or exceeding measured value that sonar to measure scope produces (if note sonar data is (x, y, θ, ra), wherein (x, y) represents the coordinate of target, θ is the orientation of the relative robot of target, and ra is that target is to the distance of robot.Undesirable data markers is ra=R, the maximum measuring distance that wherein R is sonar sensor, R=5000mm in the present invention), the data set of recording a demerit after filter is S.

Step 2, filtration singular value.In the present invention, singular value refers to not represent measured value any physical presence feature, relatively sparse.Data-oriented collection S, calculate respectively in S Euclidean distance a little between any two, and according to the size of coordinate and distance, the point in S is classified.Add up each classification and comprise number a little, when the number of classification mid point is while being less than threshold value Num, remove institute in respective class a little, Num be the threshold value of setting in advance.The data set of recording a demerit after filter singular value is S ₀, the present invention claims S ₀for range data space, referred to as metric space.

Step 3, extraction angle point.Definition length is N, and sliding step is s, and glide direction is moving window.Edge direction, from S ₀in get successively N point, remember that its transverse and longitudinal coordinate is respectively X _t=[x ₁, x ₂..., x _n] and Y _t=[y ₁, y ₂..., y _n], X _t, Y _tcovariance matrix be:

$C_t=\left[\begin{array}{cc}\mathrm{cov}(X_t,X_t)& \mathrm{cov}(X_t,Y_t)\\ \mathrm{cov}(Y_t,X_t)& \mathrm{cov}(Y_t,Y_t)\end{array}\right],---\left(1\right)$

Wherein: with be respectively X _tand Y _tthe average of element.Note C _teigenwert be λ _maxand λ _min, its ratio is EVR=λ _min/ λ _max.Along with the slip of window, can calculate S ₀all EVR values, obtain current EVR curve.At corner point, EVR can reach extreme value.The present invention calculates the peak value of current EVR curve by the mode of more adjacent EVR value, the corresponding angle point of each peak value.

Step 4, error of calculation ellipse.Calculate the peak value of current EVR curve, this step be take peak value O as example, and the computation process that elaborates error ellipse is as follows: calculate sonar data point (x corresponding to O, y), then get and take (x, y) as the center of circle, r' is n data point in radius, and its transverse and longitudinal coordinate is designated as respectively X=[x ₁..., x _n] and Y=[y ₁..., y _n].If the covariance matrix of note X, Y is C, the eigenwert of C is λ ₁and λ ₂(λ ₁>=λ ₂), corresponding proper vector is v ₁and v ₂, error ellipse is v ₁and v ₂in the coordinate system forming with for the center of circle, λ ₁and λ ₂be respectively the ellipse of major axis and minor axis, wherein

Step 5, calculating maximum error circle, set up the local map of expanding.With for the center of circle, λ ₁for radius, set up the maximum error circle that error ellipse is corresponding.According to the actual connection of angle point, calculate the public outer tangent line of the maximum error circle that the actual angle point that is connected is corresponding, further calculate the point of contact of tangent line and maximum error circle, connect the actual point of contact that connects the corresponding maximum error circle of angle point and form the local map M that expands, the center of circle of all maximum error circles, point of contact form the local crucial point set im_point that expands map M.

Step 6, map to image space.The transverse and longitudinal coordinate that note im_point comprises is a little respectively P=(p ₁..., p _m), Q=(q ₁..., q _m), p _maxand q _maxthe maximal value that represents respectively transverse and longitudinal coordinate, p _minand q _minbe respectively the minimum value of transverse and longitudinal coordinate.A bit (p in im_point _i, q _i) by formula (2) and formula (3), map to a bit (h in image space _i, k _i):

$h_i=\mathrm{int}[\mathrm{\η}(p_{\max }-p_{\min })\·\frac{(p_i-p_{\min })}{p_{\max }{-p}_{\min }}],---\left(2\right)$

$k_i=\mathrm{int}[\mathrm{\η}(q_{\max }-q_{\min })\·\frac{(q_i-q_{\min })}{q_{\max }-q_{\min }}],---\left(3\right)$

Wherein, η is scale-up factor, and η is relevant to the yardstick of sonar data and image space.

Step 7, note (h _i, k _i) be the picture in the center of circle in im_point, r _ifor corresponding radius of a circle, i=1 ..., j, j is center of circle number.With (h _i, k _i) be the center of circle, r _ifor radius, determine border circular areas, making the pixel value in the definite region of the picture at point of contact in this border circular areas and im_point is 0, otherwise is 1.Remember that the region that in this bianry image space, pixel value is 0 is the part expansion map p_M in image space.

Step 8, location matches.Location matches mainly comprises two aspects: rotational invariance coupling and two-dimensional scan coupling.

Steps A, rotational invariance coupling.

Between steps A 1, note p_M horizontal and vertical direction pixel, the maximal value of distance is respectively h _maxand v _max, the present invention defines P _c(int (h _max/ 2), int (v _max/ 2)) be the central point of p_M.With P _cfor true origin, along the horizontal and vertical direction of p_M, set up coordinate system Σ _c;

Steps A 2, make D=even ((h _max/ 2) ²+ (v _max/ 2) ²) ^1/2, " even " represents ((h _max/ 2) ²+ (v ^max/ 2) ²) ^1/2upwards get even number.At coordinate system Σ _cin, with P _cfor the center of circle, as radius, be respectively the concentric circles of r and r+ Δ r, form donut R _r.R _rthe number of pixels comprising is designated as N _r, the pixel count wherein being occupied by p_M is designated as U _r, its ratio is designated as v (r)=U _r/ N _r, be called effective duty cycle corresponding to radius r.According to this, calculate initial point to all effective duty cycles within the scope of D/2, form effective duty cycle vector V=[v (0), v (Δ r) ... v (r) ... v (D/2)] ^t.

Steps A 3, the effective duty cycle vector of establishing current and historical p_M are respectively V _l=[v _l(0), v _l(Δ r) ... v _l(r) ... v _l(D/2)] ^tand V _d=[v _d(0), v _d(Δ r) ... v _d(r) ... v _d(D/2)] ^t, V _land V _dmatching rate be:

$p_{\mathrm{rpt}}=\frac{{\mathrm{\Σ}}_{r=0}^{D/2}\left\{\right[v_l\left(r\right)-{\mathrm{\υ}}_l\left]\right[v_d\left(r\right)-{\mathrm{\υ}}_d\left]\right\}}{\sqrt{\left\{{\mathrm{\Σ}}_{r=0}^{D/2}{[v_l\left(r\right)-{\mathrm{\υ}}_l]}^2{\mathrm{\Σ}}_{r=0}^{D/2}{[v_d\left(r\right)-{\mathrm{\υ}}_d]}^2\right\}}}---\left(4\right)$

Wherein, ${\mathrm{\υ}}_l={\mathrm{\Σ}}_{r=0}^{D/2}{v}_l\left(r\right)/(1+D/2\mathrm{\Δr}),{\mathrm{\υ}}_d={\mathrm{\Σ}}_{r=0}^{D/2}{v}_d\left(r\right)/(1+D/2\mathrm{\Δr}).$

Step B, two-dimensional scan coupling.

Step B1, the establishment of coordinate system method of mentioning according to steps A 1, set up respectively the coordinate system of current p_M and historical p_M;

Step B2, remember that the pixel count of picture horizontal and vertical direction in current bianry image space is N, the pixel count of the capable and i row of i that current p_M occupies is N respectively _riand N _ci, the effective duty cycle of i row and column is designated as respectively w _ri=N _ri/ N and w _ci=N _ci/ N.Generally, due to current p_M and historical p_M towards difference, the dutycycle of same sequence number ranks is generally different.For this reason, the present invention realizes the coupling of historical p_M and current p_M according to following rule:

2.1 fixing current p_M, remember that the effective duty cycle of its i row and column is with calculate the effective duty cycle of its all ranks, generate the effective duty cycle vector of current p_M with $W_c^l=[w_{c1}^l,w_{c2}^l,...,w_{\mathrm{cN}}^l{]}^T;$

2.2 with 2.1, remember that the effective duty cycle vector of historical p_M is respectively with $W_c^d=[w_{c1}^d,w_{c2}^d,...,w_{\mathrm{cN}}^d{]}^T;$

2.3 utilize formula (4) to calculate respectively the canonical correlation coefficient of two dimensions with .Historical p_M is rotated counterclockwise 1 degree, recalculates the canonical correlation coefficient of current p_M and historical p_M, is designated as with .Successively, calculate respectively 360 groups of canonical correlation coefficients in two dimension one-periods, generate two dimensions related coefficient vector separately: with ${\mathrm{\λ}}_c={[{\mathrm{\λ}}_c^0,{\mathrm{\λ}}_c^1,...,{\mathrm{\λ}}_c^{359}]}^T.$

If the current p_M of step B3 and historical p_M represent identical local environment, related coefficient vector λ _rand λ _cthe some continuous element of middle correspondence will surpass the threshold value λ setting _th; Otherwise, λ _rand λ _celement be all less than λ _thor only have some discrete elements to surpass threshold value.Get λ _rand λ _cin comprise surpass threshold value λ _ththe average of element, be designated as respectively with the matching rate of current p_M and historical p_M

Step 9, images match.Utilize SIFT operator to extract respectively the point of interest of current p_M and historical p_M, and carry out images match, note matching rate is P _p, P in the present invention _pmake a comment or criticism really the point of interest logarithm of coupling and the ratio of all match interest point logarithms.

Step 10, matching rate merge.In the present invention, stipulate that final matching rate is P _f=α p _rpt+ β p _2d+ γ p _p, wherein α, β and γ are respectively the weight of each matching rate, meet alpha+beta+γ=1.The value of α, β and γ determines according to the matching precision of each method in practical application.If P _fdo not meet threshold value requirement, store current p_M; Otherwise, can utilize match information to carry out the tasks such as localization for Mobile Robot, map building and path planning.

The present invention has following beneficial effect

1, by sonar range data-mapping to image space, utilize image processing techniques to process sonar range data, provide a kind of new sonar data to process thinking.

2, by setting up the local map of expanding, improved the robustness that sonar data is processed.

3, proposed three kinds of sonar data image conversion matching process, three kinds of methods complement each other, and have improved matching precision.

4, the image conversion of sonar data is processed and can be effectively applied in localization for Mobile Robot, map building and path planning process, improves precision and the robustness of robot navigation's task.

Accompanying drawing explanation

Fig. 1 is sonar data image conversion Processing Algorithm process flow diagram;

Fig. 2 (a), Fig. 2 (b) are respectively data set S and S ₀, Fig. 2 (b) has shown error ellipse and maximum error circle simultaneously;

Fig. 3 (a), Fig. 3 (b) are respectively rotational invariance matching process schematic diagram and rotational invariance matching method matches result schematic diagram;

Fig. 4 (a), Fig. 4 (b) are respectively two-dimensional scan matching process schematic diagram and two-dimensional scan matching method matches result schematic diagram;

Fig. 5 (a), Fig. 5 (b) are respectively current p_M and historical p_M interest point extraction result schematic diagram;

Fig. 6 (a) is the schematic diagram of current and historical p_M SIFT interest points matching effect, and Fig. 6 (b) is that current p_M point of interest is from the schematic diagram of matching effect.

Embodiment

The present embodiment is implemented take invention technical scheme under prerequisite, provided detailed embodiment and process, but practical range of the present invention is not limited to following embodiment.

The present embodiment utilizes 16 sonar sensors image data under corridor-office environment of the Pioneer3-DX of robot equipment, utilize Visual Studio2008, OpenCV-1.0.0 and Matlab R2009a hybrid programming are realized the image conversion of sonar data and are processed, and algorithm flow as shown in Figure 1.The concrete execution step of the present invention is as follows:

(1) utilize robot sonar to gather current local environment data, the sonar data point of separated ra=5000mm, separating resulting, as shown in Fig. 2 (a), filters the singular value in sonar data, sets up metric space S ₀, as shown in Fig. 2 (b);

(2) in the present embodiment, get N=10, s=1, X _t=[x ₁, x ₂..., x ₁₀], Y _t=[y ₁, y ₂..., y ₁₀].First, by formula (1), calculate X _tand Y _tcovariance matrix C _t, then, utilize the eig () function providing in Matlab, calculate C _tproper vector and individual features value, further calculate EVR=λ _min/ λ _max.

(3) edge with step-length s moving window, be less than N to data in window.Repeating step (2), calculates the EVR that all windows are corresponding, forms current distance space S ₀eVR curve.

(4) more adjacent EVR value, calculates the peak value of current EVR curve.Suppose that O is one of them peak value, the sonar data point that O is corresponding is (x, y).Get and take (x, y) as the center of circle, the data point of the n in the circle that r' is radius, its coordinate is designated as X=[x ₁..., x _n] and Y=[y ₁..., y _n].Calculate the covariance matrix C of X, Y, the eigenvalue λ of C ₁and λ ₂and corresponding proper vector v ₁and v ₂.At v ₁and v ₂in the coordinate system forming, with for the center of circle, λ ₁and λ ₂for major axis and minor axis are determined error ellipse, wherein with for the center of circle, λ ₁for radius, determine maximum error circle.

(5) according to step (4), calculate the maximum error circle that all peak values of current EVR curve are corresponding.According to the actual connection of angle point, calculate the public outer tangent line of the maximum error circle that the actual angle point that is connected is corresponding, further calculate the point of contact of tangent line and maximum error circle, build the local map M that expands, determine the crucial point set im_point of M (as shown in Fig. 3 (a) and Fig. 4 (a)).

(6) the transverse and longitudinal coordinate that note im_point comprises is a little respectively P=(p ₁..., p _m) and Q=(q ₁..., q _m), determine p _maxand q _max, and p _minand q _min.By formula (2) and formula (3) by the point (p in im_point _i, q _i) map to the point (h in image space _i, k _i), η=0.02 in the present invention.

(7) determine (h _i, k _i), r _i, i=1 ..., j.With (h _i, k _i) be the center of circle, r _ifor radius is determined border circular areas.Order is 0 by the pixel value in the definite region of the picture at this border circular areas and im_point point of contact, otherwise is 1, determines that map p_M (as shown in Fig. 5 (a) and Fig. 5 (b)) is expanded in the part in image space.

(8) location matches.

Steps A, rotational invariance coupling.

The maximal value h of distance between steps A 1, calculating p_M horizontal and vertical direction pixel _maxand v _max, further calculate the center point P of p_M _c(int (h _max/ 2), int (v _max/ 2)), set up coordinate system Σ _c, as shown in Fig. 3 (a);

Steps A 2, first calculate D=even ((h _max/ 2) ²+ (v _max/ 2) ²) ^1/2, then calculate the effective duty cycle vector V of current and historical p_M _l=[v _l(0), v _l(Δ r) ... v _l(r) ... v _l(D/2)] ^tand V _d=[v _d(0), v _d(Δ r) ... v _d(r) ... v _d(D/2)] ^t, by formula (4), calculate V _land V _dmatching rate p _rpt(as shown in Fig. 3 (b)).

Step B, two-dimensional scan coupling.

Step B1, the establishment of coordinate system method proposing according to steps A 1, set up respectively the coordinate system of current p_M and historical p_M, as shown in Fig. 4 (a);

Step B2, calculate the effective duty cycle vector of current p_M with

Step B3, calculate the effective duty cycle vector of historical p_M with

Step B4, utilize formula (4) to calculate the canonical correlation coefficient vector of two dimensions with ${\mathrm{\λ}}_c={[{\mathrm{\λ}}_c^0,{\mathrm{\λ}}_c^1,...,{\mathrm{\λ}}_c^{359}]}^T.$

Step B5, calculating P _2d(as shown in Fig. 4 (b)).

(9) utilize SIFT operator to extract respectively the point of interest (as shown in Fig. 5 (a) and Fig. 5 (b)) of current p_M and historical p_M, and carry out images match (as shown in Fig. 6 (a) and Fig. 6 (b)), note matching rate is P _p.

(10) calculate final matching rate P _f=α p _rpt+ β p _2d+ γ p _p.In the present invention, test of many times result shows, the precision of three kinds of matching algorithms is just followed successively by: two-dimensional scan matching algorithm, and rotational invariance matching algorithm and image matching algorithm, so in this experiment, get α=0.33, β=0.37, γ=0.30.If P _fdo not meet threshold value requirement, store current p_M, otherwise, can utilize match information to carry out the tasks such as localization for Mobile Robot, map building and path planning.

Claims (11)

1. an image conversion disposal route for sonar data, is characterized in that specifically comprising the following steps:

(1) filter due to sonar blind area or exceed the measured value that sonar to measure scope produces;

(2) by the method for cluster, filter singular value and the discrete point in sonar data;

(3) calculate the EVR curve of sonar to measure value, by detecting the mode of current EVR peak of curve, obtain the environment angle point information that sonar data comprises;

(4) error ellipse of each Corner Feature that calculating sonar data comprises;

(5) calculate the maximum error circle of each error ellipse, build the current local map M that expands;

(6) set up the mapping relations in the current local map M of expansion and bianry image space;

(7) M is mapped to bianry image space, map p_M is expanded in the part in design of graphics image space;

(8) utilize rotational invariance matching process and two-dimensional scan matching process to realize the local coupling of expanding map;

(9) utilize SIFT operator extraction point of interest, and carry out images match;

(10) be rotated the fusion of unchangeability matching process, two-dimensional scan matching process and image matching method matching result.

2. the method for claim 1, is characterized in that step (1) comprising: note sonar data is (x, y, θ, ra), (x wherein, y) be coordinates of targets, θ is the orientation of the relative robot of target, and ra is that target is to the distance of robot, due to the impact of measuring blind area or exceeding sonar to measure scope, can produce undesirable data, be labeled as ra=R, the maximum measuring distance that wherein R is sonar sensor, the point of removing all ra=R, the data set after filtration is designated as S.

3. the method for claim 1, it is characterized in that step (2) comprising: calculate in sonar data collection S Euclidean distance a little between any two, and according to the size of coordinate and distance, the point in S is classified, add up each classification and comprise number a little, when the number of certain classification mid point is less than threshold value Num, remove institute in respective class a little, left point forms metric space S ₀.

4. the method for claim 1, is characterized in that step (3) comprising: definition length is N, and sliding step is s, and glide direction is moving window, edge direction, from S ₀in get successively N point, its coordinate is designated as respectively X _t=[x ₁, x ₂..., x _n] and Y _t=[y ₁, y ₂..., y _n], by:

$C_t=\left[\begin{array}{cc}\mathrm{cov}(X_t,X_t)& \mathrm{cov}(X_t,Y_t)\\ \mathrm{cov}(Y_t,X_t)& \mathrm{cov}(Y_t,Y_t)\end{array}\right]$

Calculate X _tand Y _tcovariance matrix C _t, wherein with be respectively X _tand Y _tthe average of element, note C _tthe ratio EVR=λ of eigenwert _min/ λ _max, edge with step-length s moving window, be less than N to data in window, calculate current EVR curve, by the mode of more adjacent EVR value, calculate the peak value of current EVR curve, the corresponding angle point of each peak value.

5. the method for claim 1, is characterized in that the described error of calculation ellipse of step (4) specifically comprises following content:

Suppose that O is one of them peak value, the sonar data point that O is corresponding is (x, y), get and take (x, y) as the center of circle, and the data point of the n in the circle that r' is radius, its coordinate is designated as X=[x ₁..., x _n] and Y=[y ₁..., y _n], the covariance matrix C of calculating X and Y, the eigenwert of C is λ ₁and λ ₂(λ ₁>=λ ₂), corresponding proper vector is v ₁and v ₂, error ellipse is v ₁and v ₂in the coordinate system forming with for the center of circle, λ ₁, λ ₂be respectively the ellipse of major axis and minor axis, wherein

6. the method for claim 1, is characterized in that the current local map M that expands of the described structure of step (5), specifically comprises following content:

With for the center of circle, λ ₁for radius, build the maximum error circle that error ellipse is corresponding, according to the actual connection of angle point, calculate the public outer tangent line of the maximum error circle that the actual angle point that is connected is corresponding, further calculate the point of contact of tangent line and maximum error circle, the point of contact that connects the corresponding maximum error circle of the actual angle point that is connected forms the local map M that expands, and the center of circle of all maximum error circles, point of contact form the local crucial point set im_point that expands map M.

7. the method for claim 1, is characterized in that the described mapping relations of setting up M and bianry image space of step (6) comprise:

Horizontal ordinate and ordinate that note im_point comprises are a little respectively P=(p ₁..., p _m) and Q=(q ₁..., q _m), p _maxand q _maxbe respectively the maximal value of transverse and longitudinal coordinate, p _minand q _minbe respectively the minimum value of transverse and longitudinal coordinate, a bit (p in im_point _i, q _i) pass through formula $h_i=\mathrm{int}[\mathrm{\η}(p_{\max }-p_{\min })\·\frac{(p_i-p_{\min })}{p_{\max }{-p}_{\min }}]$ With $k_i=\mathrm{int}[\mathrm{\η}(q_{\max }-q_{\min })\·\frac{(q_i-q_{\min })}{q_{\max }-q_{\min }}]$ Map to a bit (h in image space _i, k _i), wherein, η is scale-up factor.

8. the method for claim 1, is characterized in that the described part in image space set up of step (7) expands map and comprise following content:

In note im_point, the picture in the center of circle and corresponding radius are respectively (h _i, k _i) and r _i, i=1 ..., j, j is center of circle number, with (h _i, k _i) be the center of circle, r _ifor radius, determine border circular areas, making the pixel value in the definite region of the picture at point of contact in this border circular areas and im_point is 0, otherwise is 1, remembers that region that in this bianry image space, pixel value is 0 is that map p_M is expanded in part in image space.

9. the method for claim 1, is characterized in that step (8) comprising:

(8.1) utilize the current p_M of rotational invariance matching method matches and historical p_M, specifically comprise following content:

(i) calculate the maximal value h of distance between p_M horizontal and vertical direction pixel _maxand v _max, further calculate the center point P of p_M _c(int (h _max/ 2), int (v _max/ 2)), along horizontal and vertical direction, set up coordinate system Σ _c, at coordinate system Σ _cin, with P _cfor the center of circle, as radius, be respectively the concentric circles of r and r+ Δ r, form donut R _r, R _rthe number of pixels comprising is designated as N _r, the pixel count wherein being occupied by p_M is designated as U _r, its ratio is designated as v (r)=U _r/ N _r, be called effective duty cycle corresponding to radius r;

(ii) calculate D=even ((h _max/ 2) ²+ (v _max/ 2) ²) ^1/2, " even " represents ((h _max/ 2) ²+ (v _max/ 2) ²) ^1/2upwards get even number, then calculate the effective duty cycle vector V of current and historical p_M _land V _d, and pass through calculate V _land V _dmatching rate p _rpt, wherein, V _l=[v _l(0), v _l(Δ r) ... v _l(r) ... v _l(D/2)] ^t, V _d=[v _d(0), v _d(Δ r) ... v _d(r) ... v _d(D/2)] ^t, ${\mathrm{\υ}}_l={\mathrm{\Σ}}_{r=0}^{D/2}{v}_l\left(r\right)/(1+D/2\mathrm{\Δr}),{\mathrm{\υ}}_d={\mathrm{\Σ}}_{r=0}^{D/2}{v}_d\left(r\right)/(1+D/2\mathrm{\Δr});$

(8.2) utilize the current p_M of two-dimensional scan matching method matches and historical p_M, specifically comprise following content:

(i), according to the establishment of coordinate system method in (8.1), set up respectively the coordinate system of current p_M and historical p_M;

(ii) pixel count of remembering picture horizontal and vertical direction in current bianry image space is N, and the pixel count of the capable and i row of the i that occupied by p_M is N respectively _riand N _ci, the effective duty cycle of i row and column is designated as respectively w _ri=N _ri/ N and w _ci=N _ci/ N, the effective duty cycle that calculates respectively all ranks of current p_M forms effective duty cycle vector with , in like manner, calculate the effective duty cycle vector of historical p_M with

(iii) calculate respectively canonical correlation coefficient and the final matching degree P2d of current and historical p_M ranks effective duty cycle vector.

10. the method for claim 1, is characterized in that step (9) comprises following content: utilize SIFT operator to extract respectively the point of interest of current p_M and historical p_M, carry out images match, calculate matching rate P _p.

11. the method for claim 1, is characterized in that step (10) comprising:

Calculate final matching rate P _f=α p _rpt+ β p _2d+ γ p _p, alpha+beta+γ=1 wherein, α, the value of β and γ is determined according to the precision of each matching process, if P _fdo not meet threshold value requirement, store current p_M, otherwise, can utilize match information to carry out localization for Mobile Robot, map building and path planning task.