SQL Server / C# How to find the DateTime a Linestring Intersected a Polygon

Question

I am building a tracking system and want to find when someone has entered or left an area (Zone or Tier). We are collecting GPS coordinates at 5 sec intervals and they are stored as both lat/long and in a geography column along with the reading date time. The geofences for Zones and Tiers are stored as polygons in a geography column. Zones can contain Tiers and Tiers can contains Sub-Tiers. In the picture, T3 is a sub-tier. If someone is in T3, they are also in T2. The very faint set of dots on the right is also an example of what will happen. People will be going in and out of Zones and Tiers all day long.

The end result is to see a activity list

Entered Zone 1 at 6:00:00 am
Entered T2 at 6:05:00 am
Entered T3 at 7:13:12 am
Exited T3 at 7:49:32 am
etc.

Is there a way to find the points where someone enters/exits a Zone or Tier (green circles in image) and get the Reading Date Time from the GPS points table?

I have used STContains to get all people in a geofence. I have thought about using this approach, order by reading date time and if I find a 0 then 1, they entered, and conversely a 1 then 0, they exited. I am thinking this is too many loops and hoping there is a better way.

I have also tried STIntersection but I cant find a way to tie the points back to the date time. If someone is standing still for more than 5 seconds, I will get two identical points and matching on coordinates doesnt work well.

It appears that the DBGeography C# class has the same functions as SQL server. Can this be done in C# instead of SQL?

score 1 · Accepted Answer · answered May 04 '18 at 23:18

You've got a host of challenges here, but I think they're all doable.

I think you've already solved the first which is to know which regions a given data point is inside of. Either STContains() or STIntersects().

The second is that you're essentially looking for clusters of time-based contiguity. Assuming that you have reliable data collection, this is also solvable. Once you've got a set of (Person, Region, Timestamp) tuples (from above), it's a Gaps and Islands problem. A toy solution is presented below:

IF OBJECT_ID('tempdb.dbo.#observations') IS NOT NULL
    DROP TABLE #observations;
IF OBJECT_ID('tempdb.dbo.#regions') IS NOT NULL
    DROP TABLE #regions;

CREATE TABLE #observations (
    ObservationID INT NOT NULL IDENTITY,
        CONSTRAINT PK_Observations PRIMARY KEY CLUSTERED (ObservationID),
    PersonID      INT NOT null,
    Point         GEOMETRY NOT null,
    TS            DATETIME2(0) NOT NULL CONSTRAINT DF_Observations_TS DEFAULT SYSUTCDATETIME()
);

CREATE TABLE #regions (
    RegionID INT NOT NULL IDENTITY,
        CONSTRAINT PK_Regions PRIMARY KEY CLUSTERED (RegionID),
    Area GEOMETRY NOT NULL
);

INSERT INTO #regions
(
    Area
)
VALUES
( geometry::STGeomFromText('POLYGON((0 0, 1 0, 1 1, 0 1, 0 0))', 0) ),
( geometry::STGeomFromText('POLYGON((0 0, 2 0, 2 2, 0 2, 0 0))', 0) ),
( geometry::STGeomFromText('POLYGON((0 0, 4 0, 4 4, 0 4, 0 0))', 0) );

INSERT INTO #observations
(
    PersonID ,
    Point ,
    TS
)
VALUES
(   1 ,
    geometry::Point(0.5, 0.5, 0) ,
    '2018-01-01 00:00:00'
),
(   1 ,
    geometry::Point(1.5, 1.5, 0) ,
    '2018-01-01 00:00:05'
),
(   1 ,
    geometry::Point(2.5, 2.5, 0) ,
    '2018-01-01 00:00:10'
),
(   1 ,
    geometry::Point(3.5, 3.5, 0) ,
    '2018-01-01 00:00:15'
),
(   1 ,
    geometry::Point(4.5, 4.5, 0) ,
    '2018-01-01 00:00:20'
),
(   1 ,
    geometry::Point(0.5, 0.5, 0) ,
    '2018-01-01 01:00:00'
),
(   1 ,
    geometry::Point(1.5, 1.5, 0) ,
    '2018-01-01 01:00:05'
),
(   1 ,
    geometry::Point(2.5, 2.5, 0) ,
    '2018-01-01 01:00:10'
),
(   1 ,
    geometry::Point(3.5, 3.5, 0) ,
    '2018-01-01 01:00:15'
),
(   1 ,
    geometry::Point(4.5, 4.5, 0) ,
    '2018-01-01 01:00:20'
),
(   2 ,
    geometry::Point(3.5, 3.5, 0) ,
    '2018-01-01 00:00:00'
),
(   2 ,
    geometry::Point(3.5, 3.5, 0) ,
    '2018-01-01 00:00:05'
),
(   2 ,
    geometry::Point(3.5, 3.5, 0) ,
    '2018-01-01 00:00:10'
),
(   2 ,
    geometry::Point(3.6, 3.6, 0) ,
    '2018-01-01 00:00:15'
),
(   2 ,
    geometry::Point(4.5, 4.5, 0) ,
    '2018-01-01 00:00:20'
);

WITH cte AS (
    SELECT o.ObservationID,
           o.PersonID ,
           o.TS ,
           r.RegionID,
           (DATEDIFF(SECOND, '2017-01-01', o.ts)/5) - ROW_NUMBER() OVER (PARTITION BY o.PersonID, r.RegionID ORDER BY o.ts) AS gid,
           DATEDIFF(SECOND, '2017-01-01', o.ts)/5 AS diff,
           ROW_NUMBER() OVER (PARTITION BY o.PersonID, r.RegionID ORDER BY o.ts) AS rn
    FROM #observations AS o
    JOIN #regions AS r
        ON o.Point.STIntersects(r.Area) = 1
    --JOIN #timestamps AS ts
    --    ON ts.TS = o.TS
)
SELECT cte.PersonID, cte.RegionID, MIN(ts), MAX(ts)
FROM cte
GROUP BY cte.PersonID ,
         cte.RegionID,
         cte.gid;

The trick (if there is one) is realizing that row_number() increments by 1 for each member in an Island and that (number of seconds)/5 should also increment by 1 for the same criteria. So the difference of them should be constant for rows that qualify as equivalent in the same Island. That gives us a convenient value to group by.

Will this also work if there are gaps in the data? I have been considering only gathering data when they are moving to save battery, network traffic and database size. — Joshua Engelbrecht, May 07 '18 at 13:50
What you're describing is a related but different problem. What I have above uses co-locality in time as the heuristic to evaluate whether points are in the same cluster or not. You'll need to change it so that it's subsequent observations for the same person instead. I'll leave that as an exercise for the reader. — Ben Thul, May 07 '18 at 14:20

SQL Server / C# How to find the DateTime a Linestring Intersected a Polygon

1 Answers1