Markus Loecher <loecher at eden.rutgers.edu> writes:
I am trying to compute a spatial density estimator where the samples are not points but lines. The usual kernel density estimators will not work very well, because of their fixed xy orientation. I am imagining adding up many "ridge" like density contributions that are aligned with the line. (instead of Gaussian blobs). Does anyone know about algorithms or maybe even R functions that would compute such a density ?
This is implemented in the R package 'spatstat' as the function 'density.psp'. It computes the convolution between the line segment pattern and a Gaussian kernel. That means that each point on each line segment contributes a Gaussian kernel. Summing these contributions over all the points on a line segment, the total contribution from a line segment is a kind of ridge surface, aligned with the line segment. The final density estimate is the sum of such contributions from each line segment. Adrian Baddeley