CN106650749A - Method for plotting right-angled building in high-resolution optical image - Google Patents

Method for plotting right-angled building in high-resolution optical image Download PDF

Info

Publication number
CN106650749A
CN106650749A CN201611009128.XA CN201611009128A CN106650749A CN 106650749 A CN106650749 A CN 106650749A CN 201611009128 A CN201611009128 A CN 201611009128A CN 106650749 A CN106650749 A CN 106650749A
Authority
CN
China
Prior art keywords
building
point
mbr
travel
principal direction
Prior art date
Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
Granted
Application number
CN201611009128.XA
Other languages
Chinese (zh)
Other versions
CN106650749B (en
Inventor
李百寿
张强
李灵芝
Current Assignee (The listed assignees may be inaccurate. Google has not performed a legal analysis and makes no representation or warranty as to the accuracy of the list.)
Guilin University of Technology
Original Assignee
Guilin University of Technology
Priority date (The priority date is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the date listed.)
Filing date
Publication date
Application filed by Guilin University of Technology filed Critical Guilin University of Technology
Priority to CN201611009128.XA priority Critical patent/CN106650749B/en
Publication of CN106650749A publication Critical patent/CN106650749A/en
Application granted granted Critical
Publication of CN106650749B publication Critical patent/CN106650749B/en
Active legal-status Critical Current
Anticipated expiration legal-status Critical

Links

Classifications

    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
    • G06V20/00—Scenes; Scene-specific elements
    • G06V20/10—Terrestrial scenes
    • G06V20/176—Urban or other man-made structures
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F18/00—Pattern recognition
    • G06F18/20—Analysing
    • G06F18/23—Clustering techniques
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F18/00—Pattern recognition
    • G06F18/20—Analysing
    • G06F18/24—Classification techniques
    • G06F18/243—Classification techniques relating to the number of classes
    • G06F18/24323—Tree-organised classifiers
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
    • G06V10/00—Arrangements for image or video recognition or understanding
    • G06V10/20—Image preprocessing
    • G06V10/26—Segmentation of patterns in the image field; Cutting or merging of image elements to establish the pattern region, e.g. clustering-based techniques; Detection of occlusion
    • G06V10/267—Segmentation of patterns in the image field; Cutting or merging of image elements to establish the pattern region, e.g. clustering-based techniques; Detection of occlusion by performing operations on regions, e.g. growing, shrinking or watersheds
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
    • G06V10/00—Arrangements for image or video recognition or understanding
    • G06V10/40—Extraction of image or video features
    • G06V10/44—Local feature extraction by analysis of parts of the pattern, e.g. by detecting edges, contours, loops, corners, strokes or intersections; Connectivity analysis, e.g. of connected components
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
    • G06V10/00—Arrangements for image or video recognition or understanding
    • G06V10/40—Extraction of image or video features
    • G06V10/44—Local feature extraction by analysis of parts of the pattern, e.g. by detecting edges, contours, loops, corners, strokes or intersections; Connectivity analysis, e.g. of connected components
    • G06V10/457—Local feature extraction by analysis of parts of the pattern, e.g. by detecting edges, contours, loops, corners, strokes or intersections; Connectivity analysis, e.g. of connected components by analysing connectivity, e.g. edge linking, connected component analysis or slices

Abstract

The invention discloses a method for plotting a right-angled building in a high-resolution optical image. The method comprises the steps of taking the right-angled building as a plotting research object, defining the main direction of the building with minimum MBR, taking the plotting area and perimeter as evaluation criteria, fitting each linear boundary of the right-angled building in combination with a minimum out-souring rectangular constraint, judging the corner rule of each linear intersection by a direction decider, and connecting the intersections into a closed polygon according to the rule, thereby realizing plotting of the right-angled building. The method sufficiently considers the right-angled characteristic of the building, is not limited to the existing contour points when the right-angled points of the building are determined, but can derive new data points to express corners, thereby improving the extraction precision, and simultaneously solving the problems of low plotting precision, insufficient robustness and poor fitting optimization strategy of the right-angled building of the optical image in the existing method.

Description

The drawing method of right angle building in a kind of high-resolution optical image
Technical field
It is more particularly to a kind of to be constrained based on coarse contour and minimum MBR the present invention relates to high score remote sensing image identification field High-resolution optical image right angle building drawing method.
Background technology
In recent years, the spatial resolution of satellite remote-sensing image has reached and has been contained in sub-meter grade, therefore high score image Abundant atural object detailed information, this provides the data source of more horn of plenty for the research of building automatic identification and drawing method. The extraction of building and shape plotting in high score remote sensing image is the key factor that following smart city builds.
The present invention is marked and drawed by a kind of high-resolution optical image right angle building based on coarse contour and minimum MBR constraints Method, is distant to optics using a kind of more accurate geometrical fit matching best approach on the basis of the classification of building coarse contour The important geological information of sense image right angle building carries out semi-automatic extraction and marks and draws with geometry, and the process of the method includes：
Combined with object-oriented first with CART classification trees and extracted construction zone in high score image, then used mathematics Morphology is post-processed, and the coarse contour information of building is then gone out using Cany operator extractions.According to the thick result extracted, enter one The geometry of right angle building is marked and drawn in step research based on the profile of building.It is right as process is plotted as using right angle building As circular treatment minimum MBR defines building principal direction geological information, using the area and girth of plotting as evaluation criterion, knot Close the constraint of minimum outsourcing rectangle to be fitted each straight border of right angle building, turning for each straight-line intersection is differentiated by direction decision-making device Angle gauge then and each intersection point is connected into the polygon of closing by rule, realizes that right angle building is marked and drawed.
The present invention takes into full account the right angle feature of building, and it is determined that building right angle electrical when be not limited to it is existing Can be to derive new data point to represent angle point in some profile points, improve extraction accuracy.The present invention solves existing The plotting precision that has an optical image right angle building in method is not high, the problem that robustness is not enough and fitting strategy is not good enough, can be with Improve the information extraction ability of the made Target such as building in high spatial resolution image.Mark and draw result area precision and girth precision Higher, faster, algorithm complexity is lower for extraction rate, more benefits realization and promotes.
The content of the invention
The purpose of invention is to provide a kind of drawing method of right angle building in high-resolution optical image, can make high score The outline identification of building and accurately extract in resolution optical image, make extraction and the shape plotting system of right angle building Change.
Concretely comprise the following steps：
(1) a suitable segmentation yardstick is selected high-resolution remote sensing image to carry out image fusion segmentation；By image It is divided into homogeneous region, it is ensured that Image Segmentation reaches high degree of optimization.Subject area during segmentation according to different scale is taken not With segmentation yardstick, when in image target scale than it is larger when then select larger scale parameter；When extracting, destination object is less When then select less partitioning parameters.
(2) sorting parameter for obtaining sample after multi-scale segmentation of remote sensing images using expert system is differentiated into rule to determine Then classifying；Determine the candidate region of building.
(3) analyze the characteristic value of sample data and determine classification thresholds；Set up the accuracy table of CART classification trees；According to SPSS In the CART Rule Informations set up, the classifying rules that software is re-established in eCognition can be read carries out building Extraction.
(4) according to morphologic expansion, burn into opening operation, the thick building letter for extracting of closed operation mathematical principle pretreatment Breath, and will express through the description of the image boundary of Morphological scale-space.
(5) using Canny operators by filtering, increase, detect that these three processes carry out the extraction of profile, it is first flat to belong to The method differentiated after cunning.
(6) minimum enclosed rectangle (MBR) for obtaining building is found, method is that profile turns clockwise around its barycenter, Identical angle is rotated every time, the angle and MBR apex coordinates of rotation is then recorded, and calculates the girth and area of MBR；So Continue afterwards to rotate until have rotated 360 ° of home positions of going back；Then area or girth are found out most from the MBR set for obtaining Little MBR, and its angle of corresponding rotation；The minimum area MBR rotate counterclockwise f degree that finally will be obtained, the rectangle for obtaining is just It is the area minimum MBR of profile.
(7) dominant direction of building is defined, orthogonal roof border is reconstructed, the first step is to determine that orthogonal edges are polygonal Principal direction, a pair of axles in direction of occupying an leading position are orthogonal and represent the orientation at orthogonal edges edge, each boundary edge Edge answers one of parallel dominant direction, and the determination of principal direction is come fixed, with side longer in minimum MBR according to minimum MBR Move towards the principal direction as building；The trend of shorter edge is used as secondary principal direction.
(8) each side of form right angle building is obtained to the distance of a pair of right-angle sides of minimum MBR, due to the border of building Represented by coordinate points, this process can be changed into and calculate the distance for constituting the point set on side to right-angle side, using the method handle of cluster Resulting distance value is divided into N classes, and each class represents the distance apart from straight line with its mean value D, and the value that N is represented is exactly to constitute The side number of right angle building；Then it is exactly that the corresponding minimum MBR sides of N number of distance value are translated into D unit to be fitted building with this Each side of thing.
(9) border of the building being fitted by step (8), obtains orthogonal straight line, calculates the friendship of each straight line Point, the angle point or the point on boundary line of building are represented with this.
(10) intersection point is arranged in sequence by " direction of travel of a point ", then will be each according to direction of travel Press being linked in sequence of specifying and form orthogonal；So as to complete the extraction of building and the plotting of shape.
(11) with the principal direction of building and secondary principal direction as reference axis, intersecting point coordinate is the origin of coordinates, sets up local and sits Mark system, newly-established coordinate system divide into four regions plane, and the region above local coordinate origin is defined as first Quadrant (0 ° -90 ° of positive principal direction rotate counterclockwise), the region positioned at the local coordinate origin left side is used as the second quadrant (just main side To 90 ° -180 ° of rotate counterclockwise), below local coordinate origin as third quadrant (positive principal direction rotate counterclockwise 180 ° -270 °), positioned at local coordinate origin right region as fourth quadrant (270 ° of positive principal direction rotate counterclockwise - 360°)。
(12) otherwise due to local coordinate origin be positioned at building intersection point, or positioned at building side on, according to right angle The mode at turning can define tetra- kinds of forms of a, b, c, d, and two direction of travel are mutually orthogonal directions in these four forms, according to building side The trend on boundary defines two kinds of forms of e, f, and both direction of travel are opposite, so the direction of travel of the origin of coordinates is divided into 6 trends.
(13) with radius as R, a border circular areas are set up in each local coordinate origin, and is intercepted out in border circular areas Profile point coordinates, by the central coordinate of circle point of the circle of curvature of the profile fragment computations profile fragment for intercepting, according to the position of central coordinate of circle Put determination direction of travel.
(14) intersection point is determined according to the position of the curvature circle-center average coordinates in local coordinate system of the profile fragment for calculating The circle of curvature of coordinate.
(15) intersection point of non-building angle point, the center of circle of the average curvature circle of its region Internal periphery fragment is located at the intersection point office Fluctuate on a pair of principal direction axles of portion's coordinate system or in the range of very little, if be located on secondary principal direction axle (or near), its Direction of travel is principal direction trend, and its direction of travel is secondary principal direction if on the principal direction axis (or near)；If the center of circle Fall within the direction of travel that the interior then fixed direction of travel of the corresponding quadrant of point is the quadrant；Can be corresponding average according to each intersection point The home position of the circle of curvature determines the direction of travel of an intersection point.
(16) according to the direction of travel of each angle point, this point is connected in turn to form the crossing polygon of a closing, Obtain the building based on minimum MBR drawing methods and mark and draw result.
The present invention has taken into full account that the corner point of building is the feature of the principal direction of right angle and building, therefore relative to For least-square fitting approach, the precision of building plotting is significantly increased；Combined with object-oriented with CART decision trees Method realize Building extraction in eCognition, and programmed etc. using Matlab and C Plus Plus and realize proposition of the present invention The plotting that completes to right angle building shape of the methods marked and drawed of the minimum MBR based on building；Propose the walking of a point The concept in direction, unordered intersecting point coordinate is arranged according to certain order.
Description of the drawings
Fig. 1 is flow chart of the present invention.
Fig. 2 builds schematic diagram for minimum MBR of the invention.
Fig. 3 is distance measure mode of the profile coordinate of the present invention to MBR to right-angle side.
Fig. 4 is based on minimum MBR fitting a straight line schematic diagrams for the present invention.
Specific embodiment
In order to be illustrated more clearly that technical scheme, the present invention is done specifically with reference to specific embodiment Bright, following examples contribute to those skilled in the art and are better understood from the present invention.It should be pointed out that its in the art His technical staff is developing and other next examples without departing from the basis of the present invention and example, belongs to the protection of the present invention Scope.
A kind of high-resolution optical image right angle building plotting side constrained based on coarse contour and minimum MBR shown in Fig. 1 The techniqueflow chart of method, its main detailed step is as follows：
(1) spectrum of different buildings, texture, geometry, up and down will be record on airborne and Satellite-borne Detector photo-sensitive cell Text, semantic feature information, comparatively the geometric properties of building are most significant features.Building roof outline line, roof The mutual angle in outline line edge, roof area.Can slightly be taken turns using unstable spectrum, texture feature extraction building Exterior feature, is classified during this realization using context, semantic feature, and the final geometrical rule with building is crucial special Levy the accurate contour of building of extraction.
High-resolution remote sensing image is carried out into image fusion segmentation, image fusion segmentation by Object--oriented method It is that several meaningful regions, each sub-regions table are divided into different imaged objects according to a characteristic set Show a homogeneous region, while the homogeneity segmentation for requiring whole subpictures reaches the degree of height optimization.With dividing for remote sensing image The information such as the raising of resolution, texture, geometry, the context of image are more enriched, and for multi-scale division more preferable condition is provided.Choosing A certain amount of, representational data are selected.From eCognition derive sample data Brightness, The attribute informations such as RatioLayer2, GLCM series, Shape index.The form factor of image is a target image characteristics Mathematical description, according to the scope of the value for counting the specific form factor of a certain class the feature of target image can be effectively determined
(2) for different scale, the remote sensing image of different time resolution ratio, the parameter of multi-scale division also can change therewith Become；When same yardstick carries out different Objects extractions analyses, corresponding partitioning parameters are also different, and build for each atural object The result that segmentation of the result that vertical different segmentation image bearing layer is obtained than multiple atural objects under same yardstick is obtained is accurate, because This selects a suitable segmentation yardstick according to different imaged objects.Partitioning parameters have in eCognition：Each spectrum Weight, degree of compacting, form factor and segmentation yardstick.
(3) different segmentation yardsticks are taken according to different extracting objects when doing and splitting, when target in image Larger scale parameter is then selected when scale ratio is larger；Thinner partitioning parameters are then selected when extracting destination object and being less. Setting scale parameter when, partitioning parameters are excessive may to cause less divided, and so-called less divided is exactly that multiple destination objects are divided Cut to " homogeneity " region；Partitioning parameters excessively detailed rules and regulations may cause over-segmentation, and so-called over-segmentation is exactly a destination object Multiple homogeneous regions have been divided into it, both of these case can all affect last extraction accuracy, so in setting partitioning parameters When will according to selected destination object, select an appropriate partitioning parameters be very necessary.It it is multiple dimensioned point shown in Fig. 2 Cut effect contrast figure.
(4) one width images are made up of multiple wave bands, and different wave bands carries different information, to specific objective pair When as being extracted, if the object illustrates the target pair when the sensitivity of certain band class information is higher than other wave bands As the information on this wave band is more, therefore the weighted value of this wave band can be arranged larger.If certain wave band is to carrying The destination object for taking is dispensable, then can the weight of wave band setting smaller even 0 is unnecessary to remove Information improving segmentation quality.
(5) value of form factor changes the relation of color and shape segmentations criterion, by changing form criterion, while Define color criterion (color=1- shapes.The weights of form criterion are set to into 1, will cause to be obtained in Space Consistency Optimization.However, the value of form criterion can not be more than 0.9, this is the spectral information due to not considering image, the result object of segmentation Will be unrelated with spectral information.Conventional form factor computing formula is as follows：
Wherein A represents area, and C represents girth.
(6) degree of compacting criterion is used to optimize the imaged object related to degree of compacting.The criterion is applied to distinguish imaged object, These imaged objects are distinguished only according to relatively weak spectral contrast from not close object.Conventional compacts The computing formula of degree is as follows：
Wherein A represents area, and C represents girth.
(7) split the divided figure spot of the less image of yardstick thinner, split the more big divided figure spot of yardstick bigger.This Process mesoscale is excessively little, will cause over-segmentation, that is to say, that be originally an entire object figure spot be divided into it is multiple less Figure spot；Yardstick is excessively big, will cause less divided, that is be originally that two different objects have been divided into a mixing Figure spot.The homogeneous region that the less segmentation of form factor is obtained is more elongated, and the homogeneous region for obtaining of the bigger segmentation of form factor is more It is close to circle；When degree of compacting is bigger, similar region is more easily divided into a homogeneous region.
(8) sorting parameter for obtaining sample after multi-scale segmentation of remote sensing images using expert system is differentiated into rule to determine Then classifying, the characteristic value for analyzing sample data simultaneously determines classification thresholds.The accuracy table of CART classification trees is set up, shown in Fig. 3.Root According to the CART Rule Informations set up in SPSS, the classifying rules that software is re-established in eCognition can be read enters The extraction of row building, this nicety of grading is equal to more than 90%, and classification results are used for into the in-depth analysis of image；Less than 90%, Obscure for feature, it is difficult to the situation of onestep extraction, multiple loop iteration reconfiguration classification rule is set.
Consider that area will be bigger than normal than human interpretation with girth tree in information extraction result of the present invention, this is primarily due to If high score remote sensing image is using anon-normal photogra, the side of building, spectrum and the roof of side can be shown in image Easily side and roof are mixed during spectrum characteristic parameter is easily obscured, therefore the present invention is extracted causes what is automatically extracted for a class As a result it is bigger than normal；And girth is bigger than normal mainly what two reasons were caused, it is to utilize one that one of them is because when human interpretation Individual simple rectangle or polygon replace the true shape of building, and this has virtually been ignored as the detail section of building, another Individual reason is the ground class of non-building may to be classified as building when the present invention is extracted, or extracts result with " hair Thorn ", these can all increase the girth of building, cause to extract the precise decreasing of result.
(9) according to the thick architecture information for extracting of principles of mathematical morphology pretreatment, the morphology principle used is needed to have swollen It is swollen, corrosion, opening operation, closed operation etc..Morphological transformation expands (dilation)Two set are carried out using vectorial addition Merge.ExpansionIt is the set for being possible to vector plus sum, two operands of vectorial addition are respectively from X and B, And obtain arbitrarily possible combination.
If image X does with regard to B and remain in that after opening operation constant, it is claimed to be out with regard to B.If image X does with regard to B Remain in that constant after closed operation, then claim it to be to close with regard to B.
(10) will express through the construction zone contour description of Morphological scale-space, adopt：Border point set, parameter 3 kinds of methods are approached on border, curve.Border point set, by the profile of target border point set is expressed as, and each point is without order；Ginseng Number border, by the profile of target parameter curve is expressed as, and point thereon has certain order；Curve is approached, using some geometry Primitive removes the outline line of close approximation target.
(11) realize the extraction of profile using Canny operators, Canny operators be JohnCanny proposed in 1986 it with Marr (LoG) edge detection method is similar to, belong to be it is first smooth after the method differentiated.Canny operators realize that step is led to It is first to filter, is further added by, finally detects, these three processes.
(12) in 3 kinds of method kinds, bound of parameter is more accurate, there is preferable effect.The expression way is according to the plan for setting Summed data required precision is wanted in conjunction error threshold ε and practical application, from polygon point data a subset is filtered out To represent polygon, wherein ε is expressed as the vertical range that point deviates between characteristic point line cast out.Successive point is any It is divided into multiple segments, the air line distance of all coordinate points of each segment to this segment is compared with ε and is gone if within error range Fall this point, otherwise just retain.Most typical algorithm is exactly Douglas-Pu Ke (Douglas-Peucker).This method letter Just, but more sensitive to noise spot, the error for causing is often larger.
(13) the different destination object of ordinary circumstance in image the information such as the texture of regional area, gray scale and its around its There is significant change between its atural object, a step can be viewed as.Object is often concentrated in the marginal portion of object Most information, the determination of target edges with extract identification for whole image and understanding be one it is highly important because Element, is Image Segmentation, an important evidence of object detection positioning.
(14) geometry plotting is carried out to the building after reparation, the present invention is to find building based on minimum MBR Principal direction carries out building plotting.It is that profile turns clockwise around its barycenter to obtain the conventional methods of MBR, is rotated every time Identical angle, then records the angle and MBR apex coordinates of rotation, calculates the girth and area of MBR；Then proceed to rotation Until have rotated 360 ° of home positions of going back；Then the minimum MBR of area or girth is found out from the MBR set for obtaining, and The angle of its corresponding rotation；The minimum area MBR rotate counterclockwise f degree that finally will be obtained, the rectangle for obtaining is exactly the face of profile Product minimum MBR, is shown in Fig. 2, comprises the following steps that：
1). calculate centre of form Center (p, q) of profile, the coordinate of centre of form xThe coordinate of centre of form yAnd, Center (p, q) and E1(x1,y1) line and x-axis angle α1；
2). find out Ei(xi,yi) in xiAnd yiMinimum and maximum value minx, miny, maxx, maxy, with (minx, miny), (maxx, maxy), (minx, maxy), (maxx, miny) determines a rectangle as four summits of boundary rectangle；
3). calculate rectangular area A1, by A1Be assigned to MinArea, apex coordinate be assigned to Rect=(minx, miny, maxx, maxy)；
4). with Center (p, q) for pivot rotate counterclockwise profile, if every time the anglec of rotation is θ,It is E to be calculated the wide new coordinate of pivoting rear wheeli'(x'i,y'i), x'=p-cos (α1+β)*x+sin(α1 +β)*y；Y '=q-sin (α1+β)*x+cos(α1+β)*y；
5). invoked procedure 2), calculate the area A of the boundary rectangle after rotation βi, judge if (Ai＜=MinArea)；If True, then MinArea=Ai, while the apex coordinate of rectangle four of acquisition is turned clockwise β, update Rect=(min x ', min y′,m a xx′,m a yx′)；
6). repetition is called 4) and 5) until EP (end of program)；
Output：Rectangle apex coordinate collection Rect=(min x', min y', max x', max y'), rectangular area MinArea
(15) determine the principal direction of building, reconstruct orthogonal roof border, the first step is to determine that orthogonal edges are polygonal Principal direction.Occupy an leading position direction a pair of axles it is orthogonal, and represent the orientation at orthogonal edges edge, each boundary edge Answer one of parallel principal direction.The angle that angle on the prevailing direction in roof is obtained by rotate counterclockwise x-axis is defining. The orientation on roof is considered a local coordinate system (i.e. the roof system of x ' and y ' axial coordinates).The determination of principal direction is According to minimum MBR come fixed, with the trend on side longer in minimum MBR as building principal direction；Shorter edge move towards make For secondary principal direction.
(16) right angle building also have an important feature, constitute right angle building each side be parallel to corresponding to it A pair of principal directions.Can ensure that each adjacent boundary of the building of plotting is orthogonal using this feature.
(17) principal direction (Dominant Directions) is calculated：Input：Boundary rectangle vertex set Ei'(x'i,y'i)
1). choose first three point E1', E2', E3' the right-angle side for constituting, calculates the length of two right-angle sides
2). judge the length on two sides, if L1 ＞=L2, then the direction of L1 is exactly principal direction, and the direction of L2 is secondary main side To being respectively ± 90 ° of θ, θ with the angle of x-axis.
(18) segmentation of profile first calculates profile point and selects to the distance of the mutually perpendicular straight lines of minimum MBR wherein two with fitting Take contour of building to mark and draw as an example.The area minimum MBR of building is obtained by the algorithm of (14), minimum MBR is by straight line L12, L23, L34, L41 are constituted, and choose orthogonal side L12, L23.According to the range formula of point to straight line：
Calculate coordinate points E on profilei(xi,yi) to the distance of straight line L12, L23
(19) it is segmented profile using distance.The present invention use according to cluster Wave crest and wave trough point realize profile minute Section.When certain second difference score value put is more than 0, then the point represents crest；The second differential value put when certain less than 0, then the point Represent trough.According to the situation of Wave crest and wave trough aggregation, and in the distribution on level, calculate L12 and L23 translations away from From situation.Final translation distance represents with the mean value of distance in each section, sees Fig. 3.
(20) direction of travel of intersection point is determined.Obtain 7 straight lines altogether in step [19], and this 7 straight lines generation respectively Table 7 sides of building.And the intersection point of straight line then represent building angle point or building side on point.If two straight lines The linear equation of L1 and L2 is：
Try to achieve intersecting point coordinate：
7 straight line is orthogonal, has produced 12 intersection points, and the distribution situation of intersection point is as shown in Figure 4
(21) putting in order for 12 points is determined.With the principal direction of building and secondary principal direction as reference axis, intersecting point coordinate For the origin of coordinates, local coordinate system is set up, newly-established coordinate system divide into four regions plane, positioned at local coordinate origin The region of top is defined as first quartile (0 ° -90 ° of positive principal direction rotate counterclockwise), positioned at the region on the local coordinate origin left side As the second quadrant (90 ° -180 ° of positive principal direction rotate counterclockwise), below local coordinate origin as third quadrant (180 ° -270 ° of positive principal direction rotate counterclockwise), the region positioned at local coordinate origin right is used as fourth quadrant (positive principal direction 270 ° -360 ° of rotate counterclockwise).
(22) otherwise due to local coordinate origin be positioned at building intersection point, or positioned at building side on, according to right angle The mode at turning can define tetra- kinds of forms of a, b, c, d, and two direction of travel are mutually orthogonal directions in these four forms, according to building The trend on border defines two kinds of forms of e, f, and both direction of travel are opposite, so the direction of travel point of the origin of coordinates Into 6 trends.
(23) with radius as R, a border circular areas are set up in each local coordinate origin, and is intercepted out in border circular areas Profile point coordinates, by the central coordinate of circle point of the circle of curvature of the profile fragment computations profile fragment for intercepting, according to the position of central coordinate of circle Put determination direction of travel.The direction of travel of the final each point for obtaining.
(24) friendship determined according to the position of the curvature circle-center average coordinates in local coordinate system of the profile fragment for calculating The result of the point coordinates circle of curvature.From the figure, it can be seen that the intersection point of non-building angle point, the average song of its region Internal periphery fragment The center of circle of rate circle is located on a pair of principal direction axles of the intersection point local coordinate system or fluctuates in the range of very little, if being located at secondary main On axis of orientation (or near), then its direction of travel is principal direction trend, and it is walked if on principal direction axis (or nearby) Direction is secondary principal direction；The fixed direction of travel is the direction of travel of the quadrant if the center of circle is fallen within the corresponding quadrant of point. Determine that the direction of travel of an intersection point is shown in step [15] according to the home position of the corresponding average curvature circle of each intersection point.According to each angle point Direction of travel, this point be connected in turn to be formed one closing crossing polygon, here it is be based on minimum MBR plotting sides The building of method marks and draws result.
(25) summary, can be built by the detailed step of above example to high-resolution remote sensing image The shape of thing is marked and drawed.

Claims (1)

1. in a kind of high-resolution optical image right angle building drawing method, it is characterised in that concretely comprise the following steps：
(1) a suitable segmentation yardstick is selected high-resolution remote sensing image to carry out image fusion segmentation；By Image Segmentation For homogeneous region, it is ensured that Image Segmentation reaches high degree of optimization；Subject area during segmentation according to different scale takes different Segmentation yardstick, when in image target scale than it is larger when then select larger scale parameter；When extracting destination object and being less then Select less partitioning parameters；
(2) by after multi-scale segmentation of remote sensing images using expert system obtain sample sorting parameter with determine decision rule come Classification；Determine the candidate region of building；
(3) analyze the characteristic value of sample data and determine classification thresholds；Set up the accuracy table of CART classification trees；Build in foundation SPSS The CART Rule Informations for erecting, the classifying rules that software is re-established in eCognition can be read carries out carrying for building Take；
(4) according to morphologic expansion, burn into opening operation, the thick building information for extracting of closed operation mathematical principle pretreatment, and To express through the description of the image boundary of Morphological scale-space；
(5) using Canny operators by filtering, increase, detect that these three processes carry out the extraction of profile, belong to after being first smooth The method differentiated；
(6) minimum enclosed rectangle for finding acquisition building is MBR, and method is that profile turns clockwise around its barycenter, every time Identical angle is all rotated, the angle and MBR apex coordinates of rotation is then recorded, the girth and area of MBR is calculated；Then after Continuous rotation is until have rotated 360 ° of home positions of going back；Then area or girth are found out from the MBR set for obtaining minimum MBR, and its angle of corresponding rotation；The minimum area MBR rotate counterclockwise f degree that finally will be obtained, the rectangle for obtaining is exactly to take turns Wide area minimum MBR；
(7) dominant direction of building is defined, orthogonal roof border is reconstructed, the first step is to determine the polygonal main side of orthogonal edges A pair of axles to, direction of occupying an leading position are orthogonal and represent the orientation at orthogonal edges edge, and each boundary edge should One of parallel dominant direction, the determination of principal direction is come fixed, with the trend on side longer in minimum MBR according to minimum MBR As the principal direction of building；The trend of shorter edge is used as secondary principal direction；
(8) each side of form right angle building is obtained to the distance of a pair of right-angle sides of minimum MBR, the border due to building is by sitting What punctuate was represented, this process can be changed into and calculate the distance for constituting the point set on side to right-angle side, using the method for cluster gained To distance value be divided into N classes, each class represents the distance apart from straight line with its mean value D, and the value that N is represented is exactly form right angle The side number of building；Then the corresponding minimum MBR sides of N number of distance value are translated into D unit and building is fitted with this Each side；
(9) border of the building being fitted by step (8), obtains orthogonal straight line, calculates the intersection point of each straight line, with This angle point or the point on boundary line to represent building；
(10) intersection point is arranged in sequence by " direction of travel of a point ", then will be respectively pressed according to direction of travel Being linked in sequence for specifying forms orthogonal；So as to complete the extraction of building and the plotting of shape；
(11) with the principal direction of building and secondary principal direction as reference axis, intersecting point coordinate is the origin of coordinates, sets up local coordinate system, Newly-established coordinate system divide into four regions plane, and the region above local coordinate origin is defined as first quartile, Positive 0 ° -90 ° of principal direction rotate counterclockwise, the region positioned at the local coordinate origin left side is used as the second quadrant, positive principal direction inverse time Pin is rotated by 90 ° -180 °, below local coordinate origin as third quadrant, 180 ° of positive principal direction rotate counterclockwise - 270 °, the region positioned at local coordinate origin right as fourth quadrant, 270 ° -360 ° of positive principal direction rotate counterclockwise；
(12) otherwise due to local coordinate origin be positioned at building intersection point, or positioned at building side on, according to right angle corner Mode can define tetra- kinds of forms of a, b, c, d, two direction of travel are mutually orthogonal directions in these four forms, according to building border Trend defines two kinds of forms of e, f, and both direction of travel are opposite, so the direction of travel of the origin of coordinates divide into 6 Individual trend；
(13) with radius as R, a border circular areas are set up in each local coordinate origin, and intercepts out the profile in border circular areas Point coordinates, it is true according to the position of central coordinate of circle by the central coordinate of circle point of the circle of curvature of the profile fragment computations profile fragment for intercepting Determine direction of travel；
(14) intersecting point coordinate is determined according to the position of the curvature circle-center average coordinates in local coordinate system of the profile fragment for calculating The circle of curvature；
(15) intersection point of non-building angle point, the center of circle of the average curvature circle of its region Internal periphery fragment is located at the intersection point local and sits Fluctuate on a pair of principal direction axles of mark system or in the range of very little, if being located on or near secondary principal direction axle, its walking side To being principal direction trend, its direction of travel is secondary principal direction if on or near principal direction axis；If the center of circle falls within the point Then the fixed direction of travel is the direction of travel of the quadrant in corresponding quadrant；Can be according to the corresponding average curvature circle of each intersection point Home position determines the direction of travel of an intersection point；
(16) according to the direction of travel of each angle point, this point is connected in turn to form the crossing polygon of a closing, is obtained final product Result is marked and drawed to the building based on minimum MBR drawing methods.
CN201611009128.XA 2016-11-16 2016-11-16 The drawing method of right angle building in a kind of high-resolution optical image Active CN106650749B (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
CN201611009128.XA CN106650749B (en) 2016-11-16 2016-11-16 The drawing method of right angle building in a kind of high-resolution optical image

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN201611009128.XA CN106650749B (en) 2016-11-16 2016-11-16 The drawing method of right angle building in a kind of high-resolution optical image

Publications (2)

Publication Number Publication Date
CN106650749A true CN106650749A (en) 2017-05-10
CN106650749B CN106650749B (en) 2019-10-11

Family

ID=58806365

Family Applications (1)

Application Number Title Priority Date Filing Date
CN201611009128.XA Active CN106650749B (en) 2016-11-16 2016-11-16 The drawing method of right angle building in a kind of high-resolution optical image

Country Status (1)

Country Link
CN (1) CN106650749B (en)

Cited By (4)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN108304840A (en) * 2017-08-31 2018-07-20 腾讯科技（深圳）有限公司 A kind of image processing method and device
CN109903304A (en) * 2019-02-25 2019-06-18 武汉大学 A kind of contour of building automatic Extraction Algorithm based on convolutional Neural metanetwork and polygon regularization
CN111027543A (en) * 2019-11-25 2020-04-17 北京云测信息技术有限公司 Image annotation method and device
CN111639403A (en) * 2020-05-28 2020-09-08 上海理工大学 Bezier curve-based fuel rod shape optimization method

Citations (2)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US20140050368A1 (en) * 2006-05-02 2014-02-20 Geoeye Solutions, Inc. Semi-automatic extraction of linear features from image data including path width attribution
CN104915672A (en) * 2014-03-13 2015-09-16 北京大学 Rectangular building extraction method and system based on high-resolution remote sensing image

Patent Citations (2)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US20140050368A1 (en) * 2006-05-02 2014-02-20 Geoeye Solutions, Inc. Semi-automatic extraction of linear features from image data including path width attribution
CN104915672A (en) * 2014-03-13 2015-09-16 北京大学 Rectangular building extraction method and system based on high-resolution remote sensing image

Non-Patent Citations (4)

* Cited by examiner, † Cited by third party
Title
BAISHOU LI 等: "Research on the classification result and accuracy of building windows in high resolution satellite images: take the typical rural buildings in Guangxi, China, as an example", 《SPIE》 *
LIANG CHENG 等: "3D Building Model Reconstruction from Multi-view Aerial Imagery and Lidar Data", 《PHOTOGRAMMETRIC ENGINEERING & REMOTE SENSING》 *
刘春 等: "机载激光扫描中复杂建筑物轮廓线平差提取模型", 《同济大学学报（自然科学版）》 *
李百寿 等: "高分辨率遥感影像在岩墙地质体信息提取中的应用", 《地理与地理信息科学》 *

Cited By (7)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN108304840A (en) * 2017-08-31 2018-07-20 腾讯科技（深圳）有限公司 A kind of image processing method and device
CN108304840B (en) * 2017-08-31 2022-11-11 腾讯科技（深圳）有限公司 Image data processing method and device
CN109903304A (en) * 2019-02-25 2019-06-18 武汉大学 A kind of contour of building automatic Extraction Algorithm based on convolutional Neural metanetwork and polygon regularization
CN109903304B (en) * 2019-02-25 2021-03-16 武汉大学 Automatic building contour extraction algorithm based on convolutional neural network and polygon regularization
CN111027543A (en) * 2019-11-25 2020-04-17 北京云测信息技术有限公司 Image annotation method and device
CN111027543B (en) * 2019-11-25 2023-04-07 北京云测信息技术有限公司 Image annotation method and device
CN111639403A (en) * 2020-05-28 2020-09-08 上海理工大学 Bezier curve-based fuel rod shape optimization method

Also Published As

Publication number Publication date
CN106650749B (en) 2019-10-11

Similar Documents

Publication Publication Date Title
Cheng et al. Accurate urban road centerline extraction from VHR imagery via multiscale segmentation and tensor voting
CN101840581B (en) Method for extracting profile of building from satellite remote sensing image
WO2018107939A1 (en) Edge completeness-based optimal identification method for image segmentation
CN104778721B (en) The distance measurement method of conspicuousness target in a kind of binocular image
CN104008553B (en) Crack detection method with image gradient information and watershed method conflated
CN105335966B (en) Multiscale morphology image division method based on local homogeney index
CN110263717B (en) Method for determining land utilization category of street view image
CN104766046B (en) One kind is detected using traffic mark color and shape facility and recognition methods
CN104182763B (en) A kind of floristics identifying system based on flower feature
CN103971126A (en) Method and device for identifying traffic signs
CN106650749A (en) Method for plotting right-angled building in high-resolution optical image
CN103839267B (en) Building extracting method based on morphological building indexes
CN101599120A (en) A kind of identification method of remote sensing image building
CN111191628B (en) Remote sensing image earthquake damage building identification method based on decision tree and feature optimization
CN113963222B (en) High-resolution remote sensing image change detection method based on multi-strategy combination
Chen et al. Optimal segmentation of a high-resolution remote-sensing image guided by area and boundary
CN109363699A (en) A kind of method and device of breast image lesion identification
Jiao et al. A survey of road feature extraction methods from raster maps
CN109363697A (en) A kind of method and device of breast image lesion identification
CN107992856A (en) High score remote sensing building effects detection method under City scenarios
CN107292328A (en) The remote sensing image shadow Detection extracting method and system of multiple dimensioned multiple features fusion
CN109658391A (en) A kind of radius of circle measurement method being fitted based on contour mergence and convex closure
Hossain et al. A hybrid image segmentation method for building extraction from high-resolution RGB images
CN111597930A (en) Coastline extraction method based on remote sensing cloud platform
Jing et al. Island road centerline extraction based on a multiscale united feature

Legal Events

Date Code Title Description
PB01 Publication
PB01 Publication
SE01 Entry into force of request for substantive examination
SE01 Entry into force of request for substantive examination
GR01 Patent grant
GR01 Patent grant