在 UnionAggregate 中保留起点

Preserve start points in UnionAggregate

用例 1:

DECLARE @Geom TABLE 
( 
   shape geometry, 
   shapeType nvarchar(50) 
); 

INSERT INTO @Geom(shape,shapeType) 
VALUES('LINESTRING(1 2, 3 4)', 'A'), 
('LINESTRING(3.2 4, 7 8)', 'B'); 

SELECT *
FROM @Geom

SELECT geometry::UnionAggregate(shape).ToString(), geometry::UnionAggregate(shape)
FROM @Geom;

输出的 WKT 是

MULTILINESTRING ((7 8, 3.2 4), (3 4, 1 2))

什么时候我想要

MULTILINESTRING ((1 2, 3 4), (3.2 4, 7 8))

其中 "A" 和 "B" 行的开头应该是 (1 2)(3.2 4)

这种UnionAggregate的行为似乎并不关心"direction"的几何形状,以保持A union B和B union A是相同的结果。但是,我想保留 start/endpoints,因为我正在合并街道几何图形,并且我希望所有 LINESTRING 都沿其原始方向移动。

这里讨论这个问题:https://social.msdn.microsoft.com/Forums/sqlserver/en-US/89e95366-3649-4294-a0bc-f3921598157f/union-of-linestrings-and-reversing-direction?forum=sqlspatial

他们似乎建议了一个关于检查最终结果的可能解决方案,但我不清楚该怎么做。在链接的线程中暗示

The MultiLineString always represents the graph from the point which farthest from the origin point.

我不清楚这到底是什么意思,但我认为我不能假设 UnionAggregate 的结果总是与我想要的相反

如果很难知道方向意图,那么我可以添加 M 度量,其中方向应遵循增加的 M 值。

假设我有一种方法可以反转直线上的点,我将如何解决这个问题?

我发现了一个模仿 STUnion 的函数,以增加对 Z 和 M 度量的支持:http://www.spatialdbadvisor.com/files/SQLServer.html#robo48 但是注意到 "their direction could change (eg Start/Start Point relationship).",这是我想避免的。

编辑:

我还需要的功能是当 LINESTRING 有一个共享端点时,结果是连接 LINESTRING

用例 2:

DECLARE @Geom TABLE 
( 
   shape geometry, 
   shapeType nvarchar(50) 
); 

INSERT INTO @Geom(shape,shapeType) 
VALUES('LINESTRING(1 2, 3 4)', 'A'), 
('LINESTRING(3 4, 7 8)', 'B'); 

SELECT *
FROM @Geom

SELECT geometry::UnionAggregate(shape).ToString(), geometry::UnionAggregate(shape)
FROM @Geom;

这导致 WKT LINESTRING (7 8, 3 4, 1 2)

什么时候我会想要

LINESTRING (1 2, 3 4, 7 8)


尝试解决

Clay 建议的 geometry::CollectionAggregate(shape).Reduce(0) 解决了用例 1。我尝试只在结果上使用 STUnion 和一个空的线串,当它工作时它会退回到不正确的顺序。

我怀疑解决方案将是一个类似于 ST_LineMerge 的缩放函数,它获取 CollectionAggregate (MULTILINESTRING) 的结果,然后在可以合并为一个 LINESTRING 和不能合并时将这些点合并在一起returns 几何形状不变

几何类型没有 record/encode 方向性。您给它的行可能被认为是 "undirected" 或 "bi-directional"。这个returns1:

select geometry::STGeomFromText('LINESTRING(1 2, 3 4)',0).STEquals(
       geometry::STGeomFromText('LINESTRING(3 4, 1 2)',0))

因此,使用这些类型无法获得您正在寻找的内容。您认为 "start points" 很特别。我建议您将它们分别记录为个人 POINTs.

不过,这确实使所有生成的代码现在变得更难看 - 您必须将这些数据对保持在一起处理:

DECLARE @Geom TABLE 
(
   start geometry, 
   shape geometry, 
   shapeType nvarchar(50) 
); 

INSERT INTO @Geom(start,shape,shapeType) 
VALUES('POINT(1 2)','LINESTRING(1 2, 3 4)', 'A'), 
('POINT(3.2 4)','LINESTRING(3.2 4, 7 8)', 'B'); 

SELECT *
FROM @Geom

SELECT
    geometry::UnionAggregate(start).ToString(), geometry::UnionAggregate(shape).ToString(),
    geometry::UnionAggregate(start), geometry::UnionAggregate(shape)
FROM @Geom;

此时您可以决定停止直接使用地理类型 - 您可以创建一个引用 SqlGeography 的 CLR UDT(相同类型的 CLR 表面处理) 并在内部使用它,但也跟踪它的 "directionality",全部包裹在一起,然后开始使用它。

不过,您不太可能希望在该包装器中显示所有 geography 方法 - 您将不得不挑选并选择您的战斗。而且,当然,由于它并不是真正的 SQL 服务器 geography 出现在您的结果中,您将无法从 Management Studio 中的 "Spatial Results" 选项卡中受益。


在这些类型中,我唯一能想到的 "directionality" 确实存在的地方是用于消除 geography 形状歧义的左手法则。

本来我建议...

DECLARE @Geom TABLE 
( 
   shape geometry, 
   shapeType nvarchar(50) 
); 

INSERT @Geom(shape,shapeType) VALUES
  ('LINESTRING(1 2, 3 4)', 'A'), 
  ('LINESTRING(3.2 4, 7 8)', 'B'); 

SELECT * FROM @Geom

SELECT 
  geometry::CollectionAggregate(shape).Reduce(0).ToString(), 
  geometry::CollectionAggregate(shape).Reduce(0)
FROM @Geom

你得到:

...但是,我很清楚我给出的答案还不够好。例如,很难让 Reduce() 不简化你的部分台词,

我仍然喜欢 CollectionAggregate,因为它可以将您的原始线条数组合并为一个东西,但后来我认为必须有一种方法来构建必要的几何结构。

我玩了几次,这次迭代将评估为 LineStringMultiLineString,具体取决于输入中是否有不相交的 LineString 元素:

create function dbo.SimplifyToLine( @geo geometry ) returns geometry as
begin
  declare 
    @numSubGeos int = @geo.STNumGeometries(),
    @subGeoIdx int = 1,
    @sql nvarchar( max ) = N'',
    @subGeo geometry,
    @oldEndX float = -1.0e26,
    @oldEndY float = -1.0e26,
    @startX float,
    @startY float,
    @endX float,
    @endY float,
    @idx int,
    @numPoints int,
    @point geometry,
    @segment int = 1,
    @continue bit,
    @result geometry,
    @started bit = 0

  declare
    @geos table
    ( 
      Idx int primary key, 
      SubGeo geometry, 
      StartX decimal, 
      EndX decimal, 
      StartY decimal, 
      EndY decimal, 
      NumPoints int, 
      ContinueFromPrevious bit 
    ) 

  declare
    @multiLines table
    (
      Idx int primary key,
      Segment nvarchar(max)
    )

  --> collect geometries and extents...
  while ( @subGeoIdx <= @numSubGeos )  
  begin

    select @subGeo = @geo.STGeometryN( @subGeoIdx )

    select 
      @startX = @subGeo.STPointN( 1 ).STX,
      @startY = @subGeo.STPointN( 1 ).STY,
      @endX = @subGeo.STPointN( @subGeo.STNumPoints( ) ).STX,
      @endY = @subGeo.STPointN( @subGeo.STNumPoints( ) ).STY

    insert @geos values
    ( 
      @subGeoIdx,
      @subGeo, 
      @startX, 
      @endX, 
      @startY, 
      @endY, 
      @subGeo.STNumPoints() ,
      case when @subGeoIdx = 1 then 1 when @oldEndX = @startX and @oldEndY = @startY then 1 else 0 end
    )   

    select 
      @oldEndX = @endX, 
      @oldEndY = @endY, 
      @subGeoIdx = @subGeoIdx + 1
  end


  if not exists ( select * from @geos where ContinueFromPrevious = 0 ) --> then all LineStrings are connected 
  begin
    --> build a single LINESTRING( )...
    select @sql = ''
    declare c cursor for select SubGeo, StartX, EndX, StartY, EndY, NumPoints, ContinueFromPrevious from @geos order by Idx  
    open c
    while ( 1 = 1 )
    begin
      fetch next from c into @subGeo, @startX, @endX, @startY, @endY, @numPoints, @continue
      if @@fetch_status != 0 break;
      select @idx = case when @started = 0 then 1 else 2 end, @started = 1  --> accrue all points, de-duplicating line ends...
      while ( @idx <= @numPoints )
      begin
        select @point = @subGeo.STPointN( @idx )
        select @sql += convert( nvarchar, @point.STX ) + N' ' + convert( nvarchar, @point.STY ) + N','
        select @idx = @idx + 1
      end
    end
    close c
    deallocate c
    select @sql = substring( @sql, 1, len( @sql ) -1 )
    select @result =  geometry::STGeomFromText(N'LINESTRING(' + @sql + N')', 0 ) 
  end
  else  --> we have disjoint lines in the inputs...
  begin
    select @sql = N'', @started = 0
    --> build a MULTILINESTRING((),()...) with line segements terminated at disjoint points..
    declare c cursor for select SubGeo, StartX, EndX, StartY, EndY, NumPoints, ContinueFromPrevious from @geos  order by Idx
    open c
      while ( 1=1 )
      begin
        fetch next from c into @subGeo, @startX, @endX, @startY, @endY, @numPoints, @continue
        if @@fetch_status != 0 break;
        if @continue = 1
        begin
          select @idx = case when @started = 0 then 1 else 2 end, @started = 1
          while ( @idx <= @numPoints )
          begin
            select @point = @subGeo.STPointN( @idx )
            select @sql += convert( nvarchar, @point.STX ) + N' ' + convert( nvarchar, @point.STY ) + N','
            select @idx = @idx + 1
          end
        end
        else
        begin
          insert @multiLines values ( @segment, substring( @sql, 1, len( @sql ) -1 ) ) --> collect the segment
          select @idx = 1, @sql = N'', @segment = @segment + 1
          while ( @idx <= @numPoints )
          begin
            select @point = @subGeo.STPointN( @idx )
            select @sql += convert( nvarchar, @point.STX ) + N' ' + convert( nvarchar, @point.STY ) + N','
            select @idx = @idx + 1
          end
        end
      end
    close c
    deallocate c
    insert @multiLines values ( @segment, substring( @sql, 1, len( @sql ) -1 ) )
    select @sql = N''
    select @sql += N'(' + Segment + N'),' from @multiLines order by Idx --> appends all segments
    select @sql = substring( @sql, 1, len( @sql ) -1 )
    select @result = geometry::STGeomFromText( 'MULTILINESTRING('+ @sql + N')', 1 )
  end

...最后给出:

DECLARE @Geom TABLE 
( 
   shape geometry, 
   shapeType nvarchar(50) 
); 

INSERT @Geom(shape,shapeType) VALUES
  ('LINESTRING(1 2, 3 4)', 'A'), 
  ('LINESTRING(3 4, 9 9)', 'B'),  --> disjoint from here to the next LINESTRING
  ('LINESTRING(9 8, 3 4)', 'C'),
  ('LINESTRING(3 4, 1 2)', 'D'); 

select 
  dbo.SimplifyToLine(geometry::CollectionAggregate(shape)).ToString(),
  dbo.SimplifyToLine(geometry::CollectionAggregate(shape))
from 
  @Geom

delete @Geom

INSERT @Geom(shape,shapeType) VALUES
('LINESTRING(1 2, 3 4)', 'A'), 
('LINESTRING(3 4, 9 8)', 'B'),
('LINESTRING(9 8, 3 4)', 'C'),
('LINESTRING(3 4, 1 2)', 'D'); 

select 
  dbo.SimplifyToLine(geometry::CollectionAggregate(shape)).ToString(),
  dbo.SimplifyToLine(geometry::CollectionAggregate(shape))
from
  @Geom

...你得到:

与 Clay 传递 GeometryCollection 的想法不同,我实现了一个强大的版本,它将采用 POINT、MULTIPOINT、LINESTRING、MULTILINESTRING 的任意组合,并删除 @Tolerance 中的任何接触端点并创建一个 POINT、LINESTRING , 多行

这是它的工作演示(注意 0 和 0.1 的公差如何影响第二和第三个输出):

DECLARE @GeometryCollection GEOMETRY = GEOMETRY::STGeomFromText('GEOMETRYCOLLECTION (LINESTRING (1 2, 3 4), LINESTRING (3 4, 100 100), LINESTRING (9 8, 3 4), LINESTRING (3 4, 1 2), POINT(1 2), POINT(1 2), POINT(1 2))',0)
SELECT [dbo].[fnSimplifyToLine](@GeometryCollection, 0).ToString();
--Output: MULTILINESTRING ((1 2, 3 4, 100 100), (9 8, 3 4, 1 2))

SET @GeometryCollection  = GEOMETRY::STGeomFromText('GEOMETRYCOLLECTION (LINESTRING (1 2, 3 4.1), LINESTRING (3 4, 9 9, 6 1))',0)
SELECT [dbo].[fnSimplifyToLine](@GeometryCollection, 0).ToString()
--Output: MULTILINESTRING ((1 2, 3 4.1), (3 4, 9 9, 6 1))

SET @GeometryCollection  = GEOMETRY::STGeomFromText('GEOMETRYCOLLECTION (LINESTRING (1 2, 3 4.1), LINESTRING (3 4, 9 9, 6 1))',0)
SELECT [dbo].[fnSimplifyToLine](@GeometryCollection, 0.1).ToString()
--Output: LINESTRING (1 2, 3 4.1, 9 9, 6 1)

SET @GeometryCollection  = GEOMETRY::STGeomFromText('GEOMETRYCOLLECTION (POINT(1 2))',0)
SELECT [dbo].[fnSimplifyToLine](@GeometryCollection, 0).ToString()
--Output: POINT (1 2)

SET @GeometryCollection  = GEOMETRY::STGeomFromText('GEOMETRYCOLLECTION (MULTIPOINT((1 2), (2 3)))',0)
SELECT [dbo].[fnSimplifyToLine](@GeometryCollection, 0).ToString()
--Output:  (1 2, 2 3)

首先,我必须创建一个递归 CTE 函数,它采用几何并提取所有点。

CREATE FUNCTION [dbo].[fnGetPoints]
(   
    @Geometry GEOMETRY
)
RETURNS TABLE 
AS
RETURN 
(

    WITH GeometryPoints(N, Point) AS ( 
        SELECT 
            CAST(1 AS DECIMAL(9,2)) as N
            ,@Geometry.STPointN(1) as Point
        UNION ALL
        SELECT 
            CAST(N + 1.0 AS DECIMAL(9,2)) as N
            ,@Geometry.STPointN(N + 1) as Point
        FROM GeometryPoints GP
        WHERE N < @Geometry.STNumPoints()  
    )

    SELECT *
    FROM GeometryPoints
)

然后我创建了一个函数,将 fnGetPoints 交叉应用到 @GeometryCollection 中的每个几何体以获得点矩阵。使用窗口函数 (LAG) 查找端点在 @Tolerance 内的位置并删除这些点。然后我做了一个 data smear 来组合它们共享端点的几何图形。

CREATE FUNCTION [dbo].[fnSimplifyToLine] (@GeometryCollection GEOMETRY, @Tolerance DECIMAL(19,10))
RETURNS GEOMETRY
AS
BEGIN
    DECLARE @PointMatrix TABLE (
        PointId INT,
        LinestringId INT,
        GeometryIndex INT,
        GeometryType varchar(100),
        PointIndex INT,
        Point GEOMETRY,
        Duplicate BIT
    );

    DECLARE @Linestrings TABLE (
        LinestringId INT,
        PointArrayStr varchar(max)
    );

    WITH CollectionGeometries(N, Geom) AS ( 
        SELECT 
            CAST(1 AS DECIMAL(9,2)) as N
            ,@GeometryCollection.STGeometryN(1) as Geom
        UNION ALL
        SELECT 
            CAST(N + 1.0 AS DECIMAL(9,2)) as N
            , @GeometryCollection.STGeometryN(N + 1) as Geom
        FROM CollectionGeometries CG
        WHERE N < @GeometryCollection.STNumGeometries()
    ), PointMatrix AS (
        SELECT 
            ROW_NUMBER() OVER(ORDER BY G.N, P.N) as PointId
            ,G.N as GeometryIndex
            ,G.Geom.STGeometryType() as GeometryType
            ,P.N as PointIndex
            ,P.Point
        FROM CollectionGeometries G
        CROSS APPLY dbo.fnGetPoints(Geom) P
    )

    INSERT INTO @PointMatrix
    SELECT 
        PointId
        ,GeometryIndex as LinestringId
        ,GeometryIndex
        ,GeometryType
        ,PointIndex
        ,Point
        ,CASE 
            WHEN 
                GeometryIndex != LAG(GeometryIndex) OVER(ORDER BY PointId)
                AND ABS(Point.STX - LAG(Point.STX) OVER(ORDER BY PointId)) <= @Tolerance
                AND ABS(Point.STY - LAG(Point.STY) OVER(ORDER BY PointId)) <= @Tolerance
            THEN 1
            ELSE 0
        END as Duplicate
    FROM PointMatrix
    OPTION (MAXRECURSION 10000)



    -- POLYGON, MULTIPOLYGON, GEOMETRYCOLLECTION, CIRCULARSTRING, COMPOUNDCURVE, CURVEPOLYGON not supported
    IF EXISTS ( SELECT * FROM @PointMatrix WHERE GeometryType NOT IN ('POINT', 'MULTIPOINT', 'LINESTRING', 'MULTILINESTRING'))
        RETURN CAST('Geometries in @GeometryCollection must all be IN (''POINT'',''MULTIPOINT'', ''LINESTRING'', ''MULTILINESTRING'')' as GEOMETRY);

    DECLARE @SRID INT = (SELECT DISTINCT Point.STSrid FROM @PointMatrix)

    UPDATE @PointMatrix
    SET LinestringId = NULL
    WHERE GeometryIndex IN (
        SELECT GeometryIndex FROM @PointMatrix WHERE Duplicate = 1
    )

    DELETE @PointMatrix
    WHERE Duplicate = 1;

    -- Data smear
    WITH Cnt AS (
        SELECT PointId, Point, LinestringId,c=COUNT(LinestringId) OVER (ORDER BY PointId)
        FROM @PointMatrix
    ), SmearedLineStringId AS (
        SELECT PointId, Point, LinestringId=MAX(LinestringId) OVER (PARTITION BY c)
        FROM Cnt
    )

    INSERT @Linestrings
    SELECT 
        LinestringId
        ,'(' + 
                STUFF((
                    SELECT ',' + CAST(Point.STX as varchar(100)) + ' ' + CAST(Point.STY as varchar(100))
                    FROM SmearedLineStringId t2
                    WHERE t1.LinestringId = t2.LinestringId 
                    ORDER BY PointId
                    FOR XML PATH ('')
                ), 1, 1, '')
        + ')' as PointArray
    FROM SmearedLineStringId t1
    GROUP BY LinestringId

    DECLARE @Type varchar(100) = CASE 
        WHEN 1 =(SELECT COUNT(*) FROM @PointMatrix) THEN
            'POINT'
        WHEN 1 =(SELECT COUNT(*) FROM @Linestrings) THEN
            'LINESTRING'
        ELSE
            'MULTILINESTRING'
    END

    DECLARE @BeginParens char(1) = '(';
    DECLARE @EndParens char(1) = ')'

    IF @Type != 'MULTILINESTRING'
    BEGIN
        SET @BeginParens = '';
        SET @EndParens = '';
    END

    DECLARE @Wkt varchar(max) = @Type + @BeginParens + 
                STUFF((
                    SELECT ',' + PointArrayStr
                    FROM @Linestrings t2
                    ORDER BY LinestringId
                    FOR XML PATH ('')
                ), 1, 1, '')
            + @EndParens

    RETURN Geometry::STGeomFromText(@Wkt, @SRID)

END
GO