Changes between Version 5 and Version 6 of GSoC/2014/ImplementationOf3DRasterFlowLine

Timestamp:

06/20/14 20:17:50 (11 years ago)

Author:

annakrat

Comment:

week 5

GSoC/2014/ImplementationOf3DRasterFlowLine

@@ -106 +106 @@
 
 Currently the module supports different combinations of input and output. Input can be vector map of points from which the flowlines are generated, or when not providing this vector, the flowlines are generated from centers of voxels taking into account skip parameter to reduce the number of flowlines. Output can be flowaccumulation or flowlines or both. Computation of flowaccumulation takes more time since the flowlines have to be generated from each voxel (although they are not stored in the output vector map if not specified).
+
+=== Week 5 ===
+I fixed the voxel traversal algorithm and integrated in the flow accumulation computation. Computing flow accumulation works in this way: with each new point of a flowline, I test if the previous point was in the same voxel (cell) and if not I add 1 to the flow accumulation array at that particular index. Then, I check if the previous and the new point voxel index differ in more than one index (direct neighbors); if yes, I apply the traversal algorithm between these two points. This gives me indices of the possibly intersected voxels. So the flow accumulation is computed precisely comparing to a brute force approach  when  I would just make the integration step small enough. The voxel traversal algorithm was properly tested.
+
+As Soeren pointed out, there will be probably some performance issues when writing to the 3D raster at random positions.
+
+Next, I will try to implement Runge-Kutta with adaptive step size.