Method for extracting outlines of common urban buildings from airborne laser point clouds

Abstract

The present invention relates to the technical field of building outline extraction, and more specifically, to a method for extracting ordinary building outlines in cities from airborne laser point clouds. The method comprises the following steps: 1. extracting an ordered boundary point set from a building point cloud using an alpha-shape algorithm; 2. extracting a primary outline; 3. dividing the ordered boundary point set into three types of point sets based on a boundary point type classification algorithm; 4. treating points within an oblique line point set and a curve point set as occluded boundary points, and employing an IMBR method to obtain the orthogonal building outline; 5. extracting the building outline corresponding to the oblique line point set to obtain an oblique line outline segment set; 6. extracting the building outline corresponding to the curve point set to obtain a curve outline segment set; and 7. merging the points to obtain the final building outline. The present invention can preferably extract building outlines.

Description

Technical Field

[0001] The present invention relates to the technical field of building outline extraction, and in particular to a method for extracting the outlines of common urban buildings using airborne laser point clouds. Background Art

[0002] Most current building outline extraction methods extract the building's straight outline and then orthogonalize it; or they leave non-orthogonal contour segments unprocessed, thereby adapting to non-orthogonal building outlines. Currently, fewer methods can handle curved contours. The iterative CD-Spline algorithm is more adaptable and can automatically select the most appropriate order of the curve equation. However, this method requires first determining the building's corner points, dividing the boundary points into multiple point sets, and then performing variable-order spline curve fitting. Inaccurate identification of the building's boundary point set can lead to incorrect point set grouping, thus affecting the accuracy of subsequent contour fitting. Summary of the Invention

[0003] The present invention provides an airborne laser point cloud urban general building outline extraction method (GBOE method), which can overcome certain or some defects of the prior art.

[0004] According to the present invention, a method for extracting the outline of ordinary buildings in an airborne laser point cloud comprises the following steps:

[0005] 1. Use alpha-shape algorithm to extract the ordered boundary point set P of the building point cloud = {P i}, P i is the boundary point;

[0006] Second, extract the primary outline outline0 according to the recursive minimum bounding rectangle algorithm;

[0007] 3. Based on the boundary point type classification algorithm, the ordered boundary point set P is divided into three types of point sets: the parallel point set {PP} that is parallel to a line segment of outline0; the oblique straight line point set {PO} that is not parallel to all line segments of outline0; and the curve point set {PC}.

[0008] Fourth, the points in the oblique line point set {PO} and the curve point set {PC} are regarded as blocked boundary points, and the building orthogonal outline outlinePP is obtained using the IMBR method;

[0009] 5. Extract the building outline corresponding to the oblique straight line point set {PO} and obtain the oblique straight line contour segment set {LS i};

[0010] 6. Extract the building outline corresponding to the curve point set {PC} and obtain the curve contour line segment set {Curvei};

[0011] 7. Merge outlinePP, {LS i}、{Curve i}, get the final outline of the building.

[0012] Preferably, in step 3, the steps of dividing the boundary points are as follows:

[0013] 1) Construct a right-handed rectangular coordinate system UV with the lower left corner vertex of outline0 as the origin and the adjacent outline0 segment as the coordinate axis.

[0014] 2) Take the point that has not been classified as the starting point. Add the starting point to the temporary point set P temp , and starts to grow to both sides. The condition for the termination of growth is that the distance from the point to be added to the U-axis is equal to P temp The difference in the mean distance to the U-axis is greater than the error threshold T w (Determined by the plane accuracy of the point cloud). temp Joined as a subset to {PP}.

[0015] 3) Repeat step 2) until all points in P are traversed.

[0016] 4) Replace the U-axis with the V-axis and repeat steps 2) to 3). At this point, the parallel point set {PP} within the boundary point set P is obtained, and the remaining points are divided into multiple point sets {PU} of unknown types.

[0017] 5) Use least squares to PU i Perform straight line fitting and obtain the fitting error e Line ; if e Line ≤T w , PU i is a set of oblique line points; otherwise, use B-Spline to PU i Perform curve fitting and obtain the fitting error e BSpline If e BSpline ≤T w , PU i is a set of curve points; otherwise, the ordered RANSAC algorithm is used to PU i Split the oblique line to obtain multiple oblique line point sets.

[0018] 6) Repeat step 5) until all subsets in {PU} are divided into oblique line point sets or curve point sets.

[0019] 7) If two adjacent curve point sets PC i and PC i+1There is only one parallel point set PP in the middle j , PP j 、PC i and PC i+1 Merge into a temporary point set P merge . Use B-Spline to calculate the temporary point set P merge Perform fitting, if the fitting error e BSpline <T w , then the merge result P merge Add {PC} as a subset. At the same time, delete PC from {PC} i and PC i+1 , delete PP from {PP} j .

[0020] Preferably, in step 6, extracting the building outline corresponding to the curve point set {PC} includes the following steps:

[0021] a) Determine the curve point set PC based on the principle of minimum perpendicular distance from the point to the corresponding line segment i The starting point p and the end point q correspond to the line segments side(p) and side(q) in outlinePP;

[0022] b) Combine the line segments side(p) and side(q) and use the B-Spline fitting method with additional line segment constraints to fit the point set PC i Perform fitting to obtain a curve segment Curve with the starting endpoint on side(p) and the ending endpoint on side(q) i , i.e. PC i 's contour line segment;

[0023] c) Repeat steps a)-b) until all subsets in {PC} are processed, and the curve contour segment set {Curve i}.

[0024] As a preference, the B-Spline fitting method with additional line segment constraints is used to fit the point set PC i The fitting is performed as follows:

[0025] Assume that n+1 control points CP={CP0,CP1,...,CP n} and a node vector U={u0,u1,...,u n}, the definition of p-order B-Spline is as follows:

[0026] Where N i,p is the p-order B-Spline basis function, control point CP i =[X i Yi ] T ;

[0027] Assume that the point set Q={Q0,Q1,...Q m} is the set of curve points to be fitted, where Q i =[x i y i ] T The coordinates of the control points of B-Spline can be determined according to the least squares principle, as shown in the following formula:

[0028] According to the least squares estimation curve point set Q={Q0,Q1,...Q m}, the starting point of the obtained B-Spline is near Q0 and the end point is Q m Nearby; Q0 ​​and Q of the curve point set m The distance to the orthogonal contour outlinePP is greater than the distance threshold T d , so B-Spline and outlinePP will not intersect;

[0029] The starting point Q0 and the end point Q of the curve point set Q m The corresponding sides in the orthogonal outlinePP are side(Q0) and side(Q m ); In order to make the B-Spline of Q consistent with side(Q0) and side(Q m ) intersect, the starting point of the B-Spline can be located on side(Q0) and the end point on side(Q m ); Since B-Spline can make the starting point coincide with the first control point and the end point coincide with the last control point, the first and last control points can be added on side(Q0) and side(Q m ) The constraints on B-Spline make it intersect with outlinePP;

[0030] The function model of the B-Spline fitting method with additional straight line constraints can be written as:

[0031] Where, is the correction amount of the control point coordinates;

[0032]

[0033]

[0034] a1, b1, c1 are the parameters of the line equation of the line segment side(Q0); a2, b2, c2 are the parameters of the line equation of the line segment side(Q m )’s linear equation parameters; and Y i 0 It is the control point CP i Approximate values ​​of coordinates;

[0035] After adding the constraint that the first and last control points are on the straight line, the first and last control points of the calculated B-Spline must be located on the line segment side(Q0) and side(Q m ); the start and end points of the B-Spline coincide with the first and last control points, so the B-Spline intersects with outlinePP.

[0036] The present invention divides the building boundary points into three different types of boundary points, processes the three different types of boundary points separately, and adopts different contour fitting methods according to different types of point sets, thereby effectively improving the accuracy of the building contour.

[0037] The invention solves the deficiency of the IMBR algorithm in using zigzag orthogonal line segments to fit the oblique straight line contour and curved contour of a building.

[0038] The present invention avoids the feature point extraction step required by the iterative CD-Spline algorithm, thereby improving extraction accuracy. The CD-Spline algorithm requires accurate feature points to obtain an accurate boundary, but accurate feature point extraction is relatively difficult. Extraction errors can lead to the mixing of straight and curved line segments for curve fitting.

[0039] The present invention inherits the advantage of the IMBR algorithm in ensuring the orthogonality of right-angle contours.

[0040] The present invention can change the distance threshold T d To adapt to the requirements of building outline extraction for topographic maps of different scales. BRIEF DESCRIPTION OF THE DRAWINGS

[0041] Figure 1 This is a flow chart of a method for extracting the outline of common urban buildings from airborne laser point clouds in Example 1;

[0042] Figure 2 This is a schematic diagram of the boundary point type classification results in Example 1;

[0043] Figure 3 Schematic diagram of the building orthogonal outline outlinePP in Example 1;

[0044] Figure 4 This is a schematic diagram of the building outline in Example 1;

[0045] Figure 5 Schematic diagram of the parallel point set in Example 1;

[0046] Figure 6 Schematic diagram of complex point set splitting in Example 1;

[0047] Figure 7 is a schematic diagram of a B-Spline that does not intersect with outlinePP in Example 1;

[0048] Figure 8 Schematic diagram of test data of a typical building in Example 1;

[0049] Figure 9 This is a schematic diagram of the boundary point division result in Example 1;

[0050] Figure 10 This is a schematic diagram of the effect of splicing the curve point set in Example 1;

[0051] Figure 11 This is a schematic diagram of the building outline extracted by the GBOE method, IMBR algorithm, and iterative CD-Spline algorithm in Example 1;

[0052] Figure 12 This is a schematic diagram of the error contour extracted by the iterative CD-Spline algorithm in Example 1;

[0053] Figure 13 This is a schematic diagram of an example of the statistics of right-angle vertices of the contour in Example 1;

[0054] Figure 14 Schematic diagram for comparison of curve fitting in Example 1. DETAILED DESCRIPTION

[0055] In order to further understand the content of the present invention, the present invention is described in detail with reference to the accompanying drawings and embodiments. It should be understood that the embodiments are merely for explanation of the present invention and are not intended to limit the present invention.

[0056] Example 1

[0057] Based on the improved minimum bounding rectangle (IMBR) algorithm, a boundary point type classification method was introduced, and a general building outline extraction (GBOE) method for urban airborne laser point clouds was proposed.

[0058] The GBOE method is used to extract the corresponding contours from the building point cloud, so the starting data of this method is the point cloud of a single building. Figure 1 As shown, the process of the GBOE method is as follows:

[0059] 1) Use the alpha-shape algorithm to extract the ordered boundary point set P of the building point cloud = {Pi}, P i is the boundary point;

[0060] 2) Extract the primary outline outline0 according to the recursive minimum bounding rectangle (RMBR) algorithm;

[0061] 3) According to the boundary point type division algorithm, the ordered boundary point set P is divided into three types of point sets, such as Figure 2 As shown: the set of parallel points parallel to a line segment of outline0 {PP}; the set of oblique straight line points that are not parallel to all line segments of outline0 {PO}; the set of curve points {PC};

[0062] Boundary point type division is to divide the ordered boundary point set P into a parallel point set {PP}, an oblique line point set {PO}, and a curve point set {PC}, providing data for subsequent classification and contour extraction. The steps of boundary point division are as follows:

[0063] a. Construct the right-handed rectangular coordinate system UV with the lower left corner vertex of outline0 as the origin and the adjacent outline0 segment as the coordinate axis.

[0064] b. Take the point that has not been classified as the starting point. Add the starting point to the temporary point set P. temp , and starts to grow to both sides. The condition for the termination of growth is that the distance from the point to be added to the U-axis is equal to P temp The difference in the mean distance to the U-axis is greater than the error threshold T w (Determined by the plane accuracy of the point cloud). temp Joined as a subset to {PP}.

[0065] c. Repeat step b until all points in P are traversed.

[0066] d. Replace U-axis with V-axis and repeat steps b-c. At this point, we get the set of parallel points {PP} within the boundary point set P, as Figure 5 As shown, the green area point set is parallel to the V-axis, the red area point set is parallel to the U-axis, and the remaining points are divided into multiple point set sets {PU} of unknown types.

[0067] e. Use least squares to PU i Perform straight line fitting and obtain the fitting error e Line If e Line ≤T w , PU i is a set of oblique line points; otherwise, use B-Spline to PUi Perform curve fitting and obtain the fitting error e BSpline If e BSpline ≤T w , PU i is a set of curve points; otherwise, the ordered RANSAC algorithm is used to PU i Split the oblique line to obtain multiple oblique line point sets.

[0068] The ordered RANSAC algorithm requires that the extracted oblique line point set be arranged in order and that no point number jumps occur. Figure 6 The black line in the middle represents the RANSAC fit, but it contains two segments of point sets. The presence of other points between the two segments causes a jump in the point set sequence, so only the point set with the larger number of points is retained. Line extraction continues for the remaining points until no more lines can be extracted.

[0069] Using the ordered RANSAC algorithm, Figure 6 The point set in is divided into three oblique line point sets. Figure 5 The unknown type point set PU2 in is split into two oblique line point sets.

[0070] f. Repeat step e until all subsets in {PU} are divided into oblique line point sets or curve point sets.

[0071] g. If two adjacent curve point sets PC i and PC i+1 There is only one parallel point set PP in the middle j , PP j 、PC i and PC i+1 Merge into a temporary point set P merge . Use B-Spline to calculate the temporary point set P merge Perform fitting, if the fitting error e BSpline <T w , then the merge result P merge Add {PC} as a subset. At the same time, delete PC from {PC} i and PC i+1 , delete PP from {PP} j .

[0072] 4) The points in the oblique line point set {PO} and the curve point set {PC} are regarded as blocked boundary points, and the IMBR method is used to obtain the building orthogonal outline outlinePP; Figure 3 As shown in the figure, the building outline uses the line segment side j Or vertex v j The subscript j represents the serial number of the line segment or vertex;

[0073] 5) Extract the building outline corresponding to the oblique line point set {PO}:

[0074] a) Determine the oblique line point set PO based on the principle of minimum perpendicular distance from a point to the corresponding line segment i The starting point p and the end point q correspond to the line segments side(p) and side(q) in outlinePP;

[0075] b) If side(p) and side(q) are the same line segment, then PO i If it is a misjudged oblique line point set, no oblique line segment fitting is performed and the process goes to step d).

[0076] c) Find PO separately i The fitted straight line i The intersection of side(p) and side(q) is the oblique straight line segment determined by these two intersection points. i The contour segment is denoted as LS i ;

[0077] d) Repeat steps a)-c) until all subsets in {PO} are processed and the oblique straight line contour segment set {LS i}.

[0078] 6) Extract the building outline corresponding to the curve point set {PC}:

[0079] a) Determine the curve point set PC based on the principle of minimum perpendicular distance from the point to the corresponding line segment i The starting point p and the end point q correspond to the line segments side(p) and side(q) in outlinePP;

[0080] b) Combine the line segments side(p) and side(q) and use the B-Spline fitting method with additional line segment constraints to fit the point set PC i Perform fitting to obtain a curve segment Curve with the starting endpoint on side(p) and the ending endpoint on side(q) i , i.e. PC i 's contour line segment;

[0081] c) Repeat steps a)-b) until all subsets in {PC} are processed, and the curve contour segment set {Curve i}.

[0082] 7) Merge outlinePP, {LS i}、{Curve i}, get the final outline of the building, such as Figure 4 shown.

[0083] Curve contour extraction with additional straight line constraints

[0084] Curve contour extraction is to extract the curve segments of the building contour based on the curve point set. This process is divided into two steps:

[0085] 1) According to the principle of minimizing the perpendicular distance from a point to the corresponding line segment, the corresponding line segments side(p) and side(q) of the starting point p and the end point q of the curve point set are obtained in the orthogonal outline outlinePP.

[0086] There are two cases when searching for corresponding line segments: a) side(p) and side(q) are adjacent edges, such as Figure 2 In the example, side(p) of the curve subset is side1, and side(q) is side2; b) side(p) and side(q) are the same edge, that is, the curve segment intersects the same edge.

[0087] 2) Based on the line segments side(p) and side(q) and the curve point set, extract the curve line segments that intersect with side(p) and side(q).

[0088] The curve segments are extracted using the B-Spline fitting method with additional straight line constraints. This method can ensure that the fitted curve must intersect with side(p) and side(q), thereby ensuring the continuity of the building outline.

[0089] Assume that n+1 control points CP={CP0,CP1,...,CP n} and a node vector U={u0,u1,...,u n}, the definition of p-order B-Spline is as follows:

[0090] Where N i,p is the p-order B-Spline basis function, control point CP i =[X i Y i ] T ;

[0091] Assume that the point set Q={Q0,Q1,...Q m} is the set of curve points to be fitted, where Q i =[x i y i ] T The coordinates of the control points of B-Spline can be determined according to the least squares principle, as shown in the following formula:

[0092] According to the least squares estimation curve point set Q={Q0,Q1,...Q m}, the starting point of the obtained B-Spline is near Q0 and the end point is Q m Nearby; Q0 ​​and Q of the curve point set m The distance to the orthogonal contour outlinePP is greater than the distance threshold T d , so B-Spline and outlinePP will not intersect, such as Figure 7 (a)

[0093] The starting point Q0 and the end point Q of the curve point set Q m The corresponding sides in the orthogonal outlinePP are side(Q0) and side(Q m ); In order to make the B-Spline of Q consistent with side(Q0) and side(Q m ) intersect, the starting point of the B-Spline can be located on side(Q0) and the end point on side(Q m ); Since B-Spline can make the starting point coincide with the first control point and the end point coincide with the last control point, the first and last control points can be added on side(Q0) and side(Q m ) The constraints on B-Spline make it intersect with outlinePP;

[0094] The function model of the B-Spline fitting method with additional straight line constraints can be written as:

[0095] Where, is the correction amount of the control point coordinates;

[0096]

[0097]

[0098] a1, b1, c1 are the parameters of the line equation of the line segment side(Q0); a2, b2, c2 are the parameters of the line equation of the line segment side(Q m )’s linear equation parameters; and Y i 0 It is the control point CP i Approximate values ​​of coordinates;

[0099] After adding the constraint that the first and last control points are on the straight line, the first and last control points of the calculated B-Spline must be located on the line segment side(Q0) and side(Q m ); the starting point and end point of B-Spline coincide with the first control point and the last control point, so B-Spline intersects with outlinePP, such as Figure 7 (b) shown.

[0100] Comparative test

[0101] In the experiment, the scale accuracy of 1:2000 is used, that is, the length of the building outline segment less than 1.0m is not represented. Therefore, the distance threshold T of the GBOE method and the IMBR algorithm is d When using the iterative CD-Spline algorithm to extract building outlines, the Douglas-Peucker distance T dist Set to 1.0m, the angle difference between adjacent points is T ang The angle is set to 30° and the limit α of the curve iteration number is set to 10%.

[0102] Test data

[0103] The test data comes from the AHN3 point cloud dataset in the Netherlands. The point density is about 10 pts / m 2 The random error of plane accuracy is 5cm, the systematic error is 8cm, the random error of elevation accuracy is 5cm, and the systematic error is 5cm.

[0104] Four typical buildings are selected from the AHN3 point cloud dataset: a right-angle building B1 where all line segments are orthogonal ( Figure 8 (a)) Oblique straight line building B2 containing two oblique straight line segments ( Figure 8 (b)) Curved building B3 containing 1 curved line segment ( Figure 8 (c)) Complex building B4, which includes three oblique straight line segments and one curved line segment ( Figure 8 (d)).

[0105] The manually extracted building outlines are used as the basis for algorithm evaluation.

[0106] The above building point cloud is used to test the effect of GBOE method in extracting building outlines, and compared with IMBR algorithm and iterative CD-Spline algorithm.

[0107] Point set type division results

[0108] The correctness of boundary point type classification directly affects the correctness of the final extracted building outline. The boundary point classification results of 4 typical buildings are shown in Figure 9 .

[0109] The outline of building B1 consists of orthogonal straight lines, the outline of building B3 contains one curve, and the outline of building B4 contains one curve and three oblique straight lines. The boundary point division results show that the boundary points of these three buildings are completely divided correctly.

[0110] There are two oblique straight lines in the outline of building B2. The boundary point division results show that the point sets of these two oblique straight lines are correctly divided ( Figure 9(b) PO0 and PO1). However, PO2 and PO3 in the point set segmentation result are parallel point sets and are mistakenly identified as oblique line point sets. This type of point set misclassification error cannot be effectively addressed in the boundary point segmentation algorithm. The only way to determine whether to perform oblique line fitting on the point set is to determine whether the orthogonal contour segments side(p) and side(q) corresponding to the point set are identical during oblique line contour segment extraction. This prevents these misclassified point sets from affecting the contour result.

[0111] The curve point set of building B3 is merged. Before the curve point set is merged, the curve point set PC0 of B3 is divided into 3 point sets ( Figure 10 (a)). PC0 and PC1 are the curve point sets, PP 10 The three point sets meet the conditions for the curve point set union, so PC0, PP 10 , PC1 and add the merged point set to the curve point set {PC}, and at the same time add the original PC0, PP 10 , PC1 is removed from the point set. The curve point set is combined to ensure the integrity of the curve point set.

[0112] Building outline results

[0113] The building outlines of four typical buildings were extracted using the GBOE method, IMBR algorithm, and iterative CD-Spline algorithm. Building outline results ( Figure 11 ) The gray points are the building point cloud, the bright purple line segments are the extracted building outlines, and the green points are the vertices of the building outlines.

[0114] Figure 11 The building outline results show that the building outline extracted by the GBOE method is consistent with the building point cloud, and the oblique straight line outline and curved line outline of the building can also be correctly represented. The IMBR algorithm can only extract the outline of the right-angled building. For the oblique straight line or curved part of the building outline, the IMBR algorithm replaces it with a zigzag outline ( Figure 11 (b2)). Compared with the IMBR algorithm, the GBOE method extracts building outlines with higher accuracy.

[0115] The iterative CD-Spline algorithm can also extract the oblique straight line contour and curved contour of the building, but there are some obvious errors. For example, a feature point of building B2 is not correct because the angle between adjacent points is less than the angle threshold T. ang The oblique straight line profile and the adjacent orthogonal straight line profile are fitted into a straight line profile ( Figure 12 (a)). One of the two feature points in the concave orthogonal straight line profile of building B3 was missed, resulting in the orthogonal straight line profile being incorrectly fitted into two oblique straight line profiles ( Figure 12(b)). The orthogonal straight line outline of building B4 is incorrectly fitted into a curved outline ( Figure 12 (c)), the root cause is also the missed extraction of feature points.

[0116] The GBOE method accurately classifies building boundary points into three categories using a boundary point classification algorithm (results and discussion are given in 3.2.2). Based on this classification, different types of contour segments are processed categorically. Given this accurate classification of boundary point types, the building contours extracted by the GBOE method are consistent with those extracted manually, without the erroneous contours seen with the iterative CD-Spline algorithm. Compared to the iterative CD-Spline algorithm, the GBOE method extracts building contours with higher accuracy.

[0117] Profile orthogonality analysis

[0118] The angles of the corners of four building outlines were counted to quantitatively evaluate the orthogonality of the building outlines extracted by three algorithms. A corner is composed of two adjacent line segments and a vertex. If the corner is a right angle, the corresponding vertex is called a right-angle vertex. The accuracy and recall of the right-angle vertices extracted by the algorithm can be used to evaluate whether the algorithm can ensure the orthogonality of the outline.

[0119] If the right-angled vertex determined by the human is extracted by the algorithm, it is said that the right-angled vertex is correctly extracted and is represented by TP (True Positive); if the right-angled vertex determined by the human is not extracted by the algorithm, it is said that the right-angled vertex is missed and is represented by FN (False Negative); if the right-angled vertex extracted by the algorithm is manually determined to be a non-right-angled vertex, it is called an incorrect extraction and is represented by FP (False Positive).

[0120] The three algorithms were used to extract the right-angle vertices of the building outline. The statistical results are shown in Figure 13 and Table 1.

[0121] Table 1 Statistics of right-angle vertices of building outlines

[0122]

[0123] The statistical results in Table 1 show that the recall rates for right-angle vertices for both the GBOE method and the IMBR algorithm are 100%, indicating that both algorithms can correctly extract right-angled building outlines. However, the IMBR algorithm's accuracy is only 55.4%. This is because the IMBR algorithm uses jagged contours instead of oblique lines and curves, resulting in a large number of incorrectly extracted right-angle vertices. The iterative CD-Spline algorithm's recall rate for right-angle vertices is only 14.6%, indicating that the vast majority of orthogonal contours are not correctly extracted.

[0124] The orthogonality of building outlines extracted by the iterative CD-Spline algorithm is poor. There are two reasons for this:

[0125] 1) Omission of feature point extraction in the boundary point set. Omission of feature point extraction will result in the boundary points belonging to multiple straight line contours being fitted as a whole, resulting in an erroneous curve contour ( Figure 12 (c)). This results in the omission of the extraction of right-angle vertices, that is, the orthogonal straight line contours are not correctly extracted.

[0126] 2) Insufficient number of boundary points involved in contour fitting. When the orthogonal contour line segments of buildings are relatively short, there are fewer boundary points on the contour line segments, and they are affected by factors such as point cloud plane errors. In this case, fitting is performed based on only a small number of boundary points. The resulting straight line segments are likely to have an angle deviation from the actual contour line segments, resulting in the angle between the extracted contour line segments not being 90° ( Figure 11 (a3))).

[0127] The GBOE method and the IMBR algorithm are based on the minimum bounding rectangle, which is the intersection and union of rectangles to obtain the building outline. Therefore, the building outlines extracted by the MBR type algorithm are necessarily orthogonal, which is the inherent advantage of this type of algorithm.

[0128] The GBOE method treats the identified curve and oblique line points as occluded points and uses the IMBR algorithm to extract building outlines. The GBOE algorithm avoids using jagged contours to fit curve and oblique line outlines, resolving the IMBR algorithm's problem of incorrectly extracting right-angle vertices.

[0129] Contour type correctness analysis

[0130] Building outline type accuracy refers to whether the line segment types of building outlines extracted by the algorithm are consistent with the actual building outline line segment types. The accuracy of building outline line segment types, recall rate, and F1-score were used to quantitatively evaluate the accuracy of building outlines extracted by the three algorithms.

[0131] The manually extracted building outlines contain a total of 95 line segments, including 88 orthogonal straight line segments, 5 oblique straight line segments, and 2 curved line segments. The higher the F1-score of the building outline type extracted by the algorithm, the more accurate the extracted building outline type is.

[0132] The statistical results of the correctness of building outline segment types are shown in Table 2.

[0133] Table 2 Correctness statistics of contour segment types

[0134]

[0135] The statistical results show that the F1-score of the building silhouette type of the GBOE method is 100.0%, which is better than the 74.1% of the IMBR algorithm and the 82.8% of the iterative CD-Spline algorithm.

[0136] The IMBR algorithm extracts jagged outlines of oblique lines or curves. This is due to the algorithm's inherent principle, which assumes that building outline segments are orthogonal. Therefore, when a building has non-perpendicular angles and oblique lines or curves, the IMBR algorithm cannot accurately extract its outline. This jagged shape of the IMBR algorithm results in the algorithm's worst accuracy for building outline types.

[0137] The boundary point type division method of the GBOE method can accurately divide the curve point set and the oblique line point set from the ordered boundary point set (see Figure 9 ), and then classifies and extracts building outlines. The GBOE method solves the problem of the IMBR algorithm representing curved and oblique straight line outlines as zigzag outlines. The GBOE method has the highest accuracy of building outline types among the three algorithms.

[0138] The iterative CD-Spline algorithm can also extract oblique straight and curved building outlines. However, this algorithm requires extracting feature points from the boundary points using the Douglas-Peucker equation and the angles between adjacent points. The accuracy of feature point identification significantly impacts subsequent contour extraction.

[0139] The curve contour results of B4 building extracted by GBOE method and iterative CD-Spline algorithm ( Figure 14 ). The GBOE method extracts the complete curved contour. The feature points extracted by the iterative CD-Spline algorithm divide the curve into four segments, of which only one is fitted as a curve, and the remaining three are fitted as oblique lines. Therefore, for curved contours, the GBOE method extracts building contour types with higher accuracy.

[0140] The above is a schematic description of the present invention and its embodiments, which is not restrictive. The drawings show only one embodiment of the present invention, and the actual structure is not limited thereto. Therefore, if a person skilled in the art is inspired by this and, without departing from the purpose of the present invention, designs a structure and embodiment similar to this technical solution without inventiveness, they shall fall within the scope of protection of the present invention.

Claims

1. A method for extracting the outlines of common urban buildings from airborne laser point clouds, characterized by: The following steps are involved:

1. Use alpha-shape algorithm to extract the ordered boundary point set P of the building point cloud = {P i }, P i is the boundary point; Second, extract the primary outline outline0; 3. Based on the boundary point type classification algorithm, the ordered boundary point set P is divided into three types of point sets: the parallel point set {PP} that is parallel to a line segment of outline0; the oblique straight line point set {PO} that is not parallel to all line segments of outline0; and the curve point set {PC}. Fourth, the points in the oblique line point set {PO} and the curve point set {PC} are regarded as blocked boundary points, and the building orthogonal outline outlinePP is obtained using the IMBR method; 5. Extract the building outline corresponding to the oblique straight line point set {PO} and obtain the oblique straight line contour segment set {LS i }; 6. Extract the building outline corresponding to the curve point set {PC} and obtain the curve contour line segment set {Curve i }; 7. Merge outlinePP, {LS i }、{Curve i }, get the final outline of the building.

2. The method for extracting the outline of ordinary buildings in cities from airborne laser point clouds according to claim 1, characterized in that: In step 3, the steps for dividing the boundary points are as follows: 1) With the lower left corner vertex of outline0 as the origin and the adjacent outline0 line segment as the coordinate axis, construct the right-handed plane rectangular coordinate system UV; 2) Take the point that has not been classified as the starting point; Add the starting point to the temporary point set P temp , and starts to grow to both sides; the condition for the termination of growth is that the distance from the point to be added to the U-axis is equal to P temp The difference in the mean distance to the U-axis is greater than the error threshold T w ;P temp Join to {PP} as a subset; 3) Repeat step 2) until all points in P are traversed; 4) Replace U-axis with V-axis and repeat steps 2) to 3); At this point, the parallel point set {PP} within the boundary point set P is obtained, and the remaining points are divided into multiple point sets {PU} of unknown types; 5) Use least squares to PU i Perform straight line fitting and obtain the fitting error e Line ; if e Line ≤T w , PU i is a set of oblique line points; otherwise, use B-Spline to PU i Perform curve fitting and obtain the fitting error e BSpline ; if e BSpline ≤T w , PU i is a set of curve points; otherwise, the ordered RANSAC algorithm is used to PU i Split the oblique line to obtain multiple oblique line point sets; 6) Repeat step 5) until all subsets in {PU} are divided into oblique line point sets or curve point sets; 7) If two adjacent curve point sets PC i and PC i+1 There is only one parallel point set PP in the middle j , PP j 、PC i and PC i+1 Merge into a temporary point set P merge ; Use B-Spline to calculate the temporary point set P merge Perform fitting, if the fitting error e BSpline <T w , then the merge result P merge Add {PC} as a subset; delete PC from {PC} i and PC i+1 , delete PP from {PP} j .

3. The method for extracting the outline of ordinary buildings in cities from airborne laser point clouds according to claim 2, characterized in that: In step 6, extracting the building outline corresponding to the curve point set {PC} includes the following steps: a) Determine the curve point set PC based on the principle of minimum perpendicular distance from the point to the corresponding line segment i The starting point p and the end point q correspond to the line segments side(p) and side(q) in outlinePP; b) Combine the line segments side(p) and side(q) and use the B-Spline fitting method with additional line segment constraints to fit the point set PC i Perform fitting to obtain the curve segment Curve with the starting endpoint on side(p) and the ending endpoint on side(q) i , i.e. PC i 's contour line segment; c) Repeat steps a)-b) until all subsets in {PC} are processed, and the curve contour segment set {Curve i }.

4. The method for extracting the outline of ordinary buildings in cities from airborne laser point clouds according to claim 3 is characterized by: B-Spline fitting method using additional line segment constraints for point set PC i The fitting is performed as follows: Given n+1 control points CP={CP0,CP1,...,CP n } and a node vector U={u0,u1,...,u n }, the definition of p-order B-Spline is as follows: Where N i,p is the p-order B-Spline basis function, control point CP i =[X i Y i ] T ; Point set Q = {Q0,Q1,...Q m } is the set of curve points to be fitted, where Q i =[x i y i ] T The coordinates of the control points of B-Spline can be determined according to the least squares principle, as shown in the following formula: According to the least squares estimation curve point set Q={Q0,Q1,...Q m }, the starting point of the obtained B-Spline is near Q0 and the end point is Q m Nearby; Q0 ​​and Q of the curve point set m The distance to the orthogonal contour outlinePP is greater than the distance threshold T d , so B-Spline and outlinePP will not intersect; The starting point Q0 and the end point Q of the curve point set Q m The corresponding sides in the orthogonal outlinePP are side(Q0) and side(Q m ); In order to make the B-Spline of Q consistent with side(Q0) and side(Q m ) intersect, the starting point of the B-Spline can be located on side(Q0) and the end point on side(Q m ); Since B-Spline can make the starting point coincide with the first control point and the end point coincide with the last control point, the first and last control points can be added on side(Q0) and side(Q m ) The constraints on B-Spline make it intersect with outlinePP; The function model of the B-Spline fitting method with additional straight line constraints can be written as: Where, is the correction amount of the control point coordinates; a1, b1, c1 are the parameters of the line equation of the line segment side(Q0); a2, b2, c2 are the parameters of the line equation of the line segment side(Q m )’s linear equation parameters; and Y i 0 It is the control point CP i Approximate values ​​of coordinates; After adding the constraint that the first and last control points are on the straight line, the first and last control points of the calculated B-Spline must be located on the line segment side(Q0) and side(Q m ); the start and end points of the B-Spline coincide with the first and last control points, so the B-Spline intersects with outlinePP.

Images