One of the most important aspects when interpolating point clouds into meshed DEM’s, is parameter impact on determinations of early segmentation between surface and terrain, and thus indirectly the amount of information present to define monuments of cultural heritage. Because, most filtration algorithms focus on removing modern anthropogenic structures, many of them sadly also remove ancient anthropogenic structures.
A filtration process could be defined as below:
1. Calculate the average height distance H between the points:
a) search nearest neighbor of point P(i)
b) measure height difference H(i) between P(i) and its nearest neighbor
c) sum up all h(i) and divide by the number of points in the circle
2. if H is higher than user defined value, lower threshold T
3. delete all points above the threshold T
4 .go to the next cell, until all cells are examined.
Filters always operate on a local neighborhood. The filtering of a neighborhood can be done in three possible ways.
1. Point-to-Point (1:1) – In these algorithms two points are compared at a time. The discriminant function is based on the positions of the two points. If the output of the discriminant function is above a certain threshold then one of the points is assumed to belong to an object. Only one point is classified at a time.
2. Point-to-Points (1:m) – In these algorithms neighboring points (of a point of interest) are used to solve a discriminant function. Based on the output of the discriminant function the point of interest can then be classified. One point is classified at a time.
3. Points-to-Points (n:m) – In these algorithms several points are used to solve a discriminant function. Based on the discriminant function the points can then be classified. More than one point is classified in such a formulation.
Optimizing object detection of hidden cultural heritage
LiDAR characteristics influencing segmentation:
- Resolution of data
- Overlapping surface
Suppose landscape is densely vegetated, or vegetation is in direct connection with structures of interest: What part of the point cloud goes to what class? Perhaps easy to distinguish for a human operator, but very difficult for an automated algorithm