Hi, I am looking for ways to improve the performance of an application using polytope intersectors for selection. Basically we are using an intersense tracker attached to a user's hand (paired with a data glove) and he is using it to select and manipulate objects. Small objects are a pain to select using line segment intersectors so I need to use polytope intersection...
Right now in my benchmark application, running a simple polytope intersector (in the form of a pyramid, with each plane being rotated by 0.5 degrees outwards), I have an FPS of about 15 when I'm pointing in the blank and that drops quite quickly when I point at geometrically complex objects. Now according to this we can gain performance by limiting the data accumulated by the polytope intersector. Peter Hrenka wrote : > > The main problem with PoltopeIntersector is that it generates large > amounts of result data if it is used with large picking volumes. > > The typical usage (as shown in osgkeyboardmouse example) is > picking small volumes where only "a handful" of picks occur. > > But you are not the first one that tries to use it with larger volumes... > > One possible optimization I can think of is changing the result > data structure in such a way that it only stores the intersecting > node-path and the first intersecting primitive. This data should > be enough for your usage pattern. The detailed result data could > be generated in an extra step (at a cost). > All I need from my Intersections is the node path and the localIntersectPoint. I see some of the information is stocked/copied in the function 'void PolytopeIntersector::intersect(osgUtil::IntersectionVisitor& iv, osg::Drawable* drawable)', can I get the performance boost just by modifying this function or do I need to delve deeper and tweak PolytopeIntersection and PolytopePrimitiveIntersector ? On the same topic but with a different approach, would it provide a significant increase in performance to calculate the distance of intersections to a reference line instead of a reference plane (less multiplications) ? I have also read on this mailing list that using a kdtree could boost the performance of the polytope intersector. Peter Amstutz said in a post ( http://www.mail-archive.com/[email protected]/msg33056.html) that he would consider implementing it. Any news on this side ? How significant a boost could we be expecting by using kdtrees ? thanks a lot! Julien-Charles
_______________________________________________ osg-users mailing list [email protected] http://lists.openscenegraph.org/listinfo.cgi/osg-users-openscenegraph.org
