sql - How to filter points based on proximity while inserting into PostGIS?

I have a large amount of locations that need to be inserted to a PostGIS database. However, I'd like to aggregate these points so that if there's already a location, say nearer than a kilometer, then to discard that point. I'd like a pointer or two on the appropriate query (or, if it can't be done in a reasonable way, then flame me...:)

I reckon it has to be done a bit like:

INSERT INTO locations VALUES (ST_GeographyFromText('POINT(54.55 26.33)')) 
IF ST_Distance(ST_GeographyFromText('POINT(54.55 26.33)'), ) > 1000;

As most of the audience has probably understood by now, I'm not that bright when it comes to databases so don't be too harsh on me...

Answer

You can use ST_Distance and convert to geography to test if there exists at least one location less than 1000 meters away:

SELECT
    COUNT(*) = 0 AS should_insert
FROM
    locations,
    (
        VALUES( SetSRID(MakePoint(54.55, 26.33), 4326) )
    ) AS new_value
WHERE

    center IS NOT NULL
AND
    ST_Distance( column1::geography, center::geography ) < 1000
LIMIT
    1
;

where 54.55, 26.33 is the value you want to test against locations.center in that example (column1 is the name automatically assigned by the usage of VALUES()).

Then depending on the boolean result, you can decide to insert or not.

---

Otherwise, a more efficient method could be to insert all locations to a temporary (or even better unlogged if you have PostgreSQL 9.1) table, then let PostGIS cluster into cells, and use the result of this clustering to insert:

INSERT INTO locations (id, center)
SELECT
    ids[1] AS id,
    centers[1] AS center
FROM
(
    SELECT
        array_agg(id) AS ids,
        array_agg(center) AS centers

    FROM temporary_table
    GROUP BY ST_SnapToGrid( ST_Transform(ST_SetSRID(center, 4326), 2163), 1000, 1000 )
) AS grouped
;

(this example works only with North American positions because of SRID 2163).