THREE.Box3.setFromObject( object ) & THREE.BoundingBoxHelper( object );

[WestLangley]

This is a proposal for a new Box3 method Box3.setFromObject( object ) that computes the world-axis-aligned bounding box of an object (including its children), accounting for both the object's, and childrens', world transforms.

This is different from Geometry.computeBoundingBox() which is not hierarchical, and does not take transforms into consideration.

THREE.Box3.prototype.setFromObject = ( function() {
    var v1 = new THREE.Vector3();
    return function( object ) {
        var scope = this;
        object.updateMatrixWorld( true );
        this.makeEmpty();
        object.traverse( function ( node ) {
            if ( node.geometry !== undefined && node.geometry.vertices !== undefined ) {
                var vertices = node.geometry.vertices;
                for ( var i = 0, il = vertices.length; i < il; i++ ) {
                    v1.copy( vertices[ i ] );
                    v1.applyMatrix4( node.matrixWorld );
                    scope.expandByPoint( v1 );
                }
            }
        } );
        return this;
    };
}());

In addition, I propose a BoundingBoxHelper ( different from the existing BoxHelper as it is axis-aligned ) to show the world-axis-aligned bounding box for an object. This helper calls Box3.setFromObject( object ).

bbHelper = new THREE.BoundingBoxHelper( object, 0xffffff );
scene.add( bbHelper );

The advantages of these features should be pretty obvious. The disadvantage is that computation of the AABB of an object and it's children requires processing each vertex individually -- there are no transforms that can be applied to just the geometry.boundingBox to get the correct result.

Feedback as to usefulness, naming, etc. is welcome. Maybe Object3D.computeBoundingBox() is better.

Fiddle: http://jsfiddle.net/MREL4/

[mrdoob]

Yes yes yes!

Uhm, somehow Object3D.computeBoundingBox() sounds better to me. Maybe Object3D could also have both boundingSphere and boundingBox initialised as null.

[gaitat]

Also prefer Object3D.computeBoundingBox()

Fixing Issue #3370 or something must be added to the collada loader in regards to the bounding box ?

After having looked at the fiddle, shouldn't the update() be happening automatically? It is part of the scene so an additional call shouldn't be required.

[WestLangley]

In the PR, I did create Object3D.computeBoundingBox(), and added a boundingBox property to the Object3D class.

I'm thinking that perhaps my original idea was better -- not having an Object3D.boundingBox property, and instead providing Box3.setFromObject( object ).

The reason is that an object's bounding box becomes invalid after every transformation, and we can't afford to recompute it each time -- like we do for geometries in Geometry.applyMatrix().

Hmmm....

[mrdoob]

Yup, I see what the issue... Your original idea sounds best to me now.

[WestLangley]

Yup, I see what the issue... Your original idea sounds best to me now.

OK. Backed out and reimplemented.

[potomek]

Can the bounding box functions compute also actual appearance of geometries, I mean morph targets interpolations which offset vertices ? :-)

[WestLangley]

@potomek No, the bounding box functions do not take into account morph targets.

[potomek]

So I will add that but not yet, because I'm still learning of three.js