Geometry Boost List: Re: [geometry] distance between geometries

Subject: Re: [geometry] distance between geometries
From: Adam Wulkiewicz (adam.wulkiewicz_at_[hidden])
Date: 2013-09-02 15:29:52

Hi,

Barend Gehrels wrote:
> On 2-9-2013 0:04, Adam Wulkiewicz wrote:
>>
>> Initially I thought about providing ray queries. Those queries return 
>> some number of objects hit by a ray.
>>
>>
>>
>>
>> Because we don't have a Ray concept I decided to use Segment and by 
>> extension Linestring. I thought about a real intersection because 
>> this algorithm will probably be the fastest one but more general 
>> variations could also be useful. I'd like to allow users using any 
>> algorithm they want in a strategy of the nearest() predicate. More 
>> algorithms means more possibilities.
>>
>> Btw, Is this what you've thought?
>>
>>
>
>
> Yes, based on your proposal, I think that intersecting and objects 
> close by are both interesting. But if you want the first 5, you might 
> get ambiguities (closer to starting point, or closer to the line...)
>

Yes, this is why I'd like to allow users define their strategies and 
provide some ready-to-use common ones. The default one would use 
bg::comparable_distance so wouldn't take into account the position on 
the line.

>> Yes, the knowledge about the position of a measure-point in some 
>> cases is useful, e.g. in the Linestring example above where the 
>> position of a closest point must be calculated and then it's distance 
>> from the beginning of the Linestring.
>>
>> I thought about something related to this and nearest predicate. If 
>> we have two points inside a Polygon or Two Polygons intersecting one 
>> Point, which one is the closest?
>>
>>
>>
>> If I remember correctly, currently I'm using distance = 0 if the 
>> Point is inside (intersecting) the Box but one could also say that 
>> the closest one is the one nearest to the centroid or further of the 
>> border or even something else.
>
> I see, so this is another sort of ambiguity indeed. I don't know what 
> users expect if they ask for the 5 nearest points and there are 10 
> inside. But using the centroid is an interesting solution. However, if 
> two points are located on the same distance (w.r.t. centroid), you can 
> still get an arbitrary choice... I don't think that is really 
> problematic, though I'm quite sure some users want to have a 
> consistent choice in these cases.
>

Exactly my way of thinking. Thanks.

Best regards,
Adam