Best parameters map.hougt

[ghost]

@tiagodc @caiohamamura

I'm new here and I'm having a lot of difficulty.

I don't work with high dot density, so I tested several parameters to try to find the best result.

the classifier often finds more than one tree at the same location. That's why I applied the treeMap.merge feint, but now it always identifies one less tree than it should.

At the same time, while setting minimum and maximum height in map.hough(<h_min, h_max, min_density>), it selects even fewer trees, so I need to reduce my density even more. (), It still doesn't identify everything.

Could you suggest an example to me so that I can better understand this process.

another thing that I find interesting is that I do the normalization, but when I ask to remove the ground points () it doesn't do it. in addition to appearing as an error message. apparently the point cloud was not normalized

[spokswinski]

There are a few options here. You could go with conservative parameters that under select trees, then manually add them using the map.pick function, or you could use nearest neighbor metrics using the fastPointMetrics function to preselect possible tree regions. I normalize, then voxelize, then apply fastPointMetrics using Verticality and Eigentropy, then filter the voxelized point cloud by high Verticality (between 80 and 95) and low eigentropy (below 0.03), then use the filtered point cloud with the treeMap function and map.hough.

If you want to get rid of the ground points, it would be: tls = tlsNormalize(tls, keep_ground = F)

Pro tip on the normalization, after normalizing, filter the points less then zero with the following (tls being the original point cloud) :

tls <- filter_poi(tls, Z > 0L)

[ghost]

@spokswinski Thank you for your contribution.

Even using the tls = tlsNormalize (tls, keep_ground = F) function, it does not remove the ground points.

Could you please share your code with me as an example?

I didn't understand how to use these parameters:

INDEX        METRIC ENABLED
1      1             N    TRUE
2      2       MinDist   FALSE
3      3       MaxDist   FALSE
4      4      MeanDist    TRUE
5      5        SdDist   FALSE
6      6     Linearity   FALSE
7      7     Planarity   FALSE
8      8    Scattering   FALSE
9      9  Omnivariance   FALSE
10    10    Anisotropy   FALSE
11    11    Eigentropy   FALSE
12    12      EigenSum   FALSE
13    13     Curvature    TRUE
14    14     KnnRadius   FALSE
15    15    KnnDensity   FALSE
16    16   Verticality    TRUE
17    17        ZRange   FALSE
18    18           ZSd   FALSE
19    19   KnnRadius2d   FALSE
20    20  KnnDensity2d   FALSE
21    21    EigenSum2d   FALSE
22    22  EigenRatio2d   FALSE
23    23   EigenValue1   FALSE
24    24   EigenValue2   FALSE
25    25   EigenValue3   FALSE
26    26 EigenVector11   FALSE
27    27 EigenVector21   FALSE
28    28 EigenVector31   FALSE
29    29 EigenVector12   FALSE
30    30 EigenVector22   FALSE
31    31 EigenVector32   FALSE
32    32 EigenVector13    TRUE
33    33 EigenVector23    TRUE
34    34 EigenVector33    TRUE

This is my code:

las = readLAS(file)
x = plot(las)

tls = tlsNormalize(las, keep_ground = F)
tls <- filter_poi (tls, Z> 0L)

thin = tlsSample(tls, smp.voxelize(spacing= 0.05))

all_metrics = fastPointMetrics.available ()
my_metrics = all_metrics [c (11,16)]
tls = fastPointMetrics(thin, ptm.knn(10),my_metrics)
map = treeMap(thin,map.hough(min_h = 1.5, max_h = 4, min_density = 0.00001), positions_only = F)
a = treeMap.merge(map, d= 0.5)
map
length(a$TreeID)
add_treeMap(x, map, color='yellow', size=4)
tls = treePoints(tls, map, trp.crop())
add_treePoints(x, tls, size=4)
add_treeIDs(x, tls, cex = 2, col='yellow')

tls = stemPoints(tls, stm.hough(min_density = 0.00001))
add_stemPoints(x, tls, color='red', size=1)

inv= tlsInventory (
  tls,
  d_method = shapeFit(shape = "circle", algorithm = "ransac")
)

seg = stemSegmentation(tls, sgt.ransac.circle(n = 15))
add_stemSegments(x, seg, color='white', fast=T)

df = data.frame(inv)
df

Thanks

[spokswinski]

Okay. We add a few things to this.

There are some points that are left below the ground as an artifact of how the ground points are classified. We use the following to get rid of that:

tls = tlsNormalize(tls, keep_ground = F) tls <- filter_poi(tls, Z > 0L)

Then, to use the two of the parameters you called with all_metrics, we pick out 16 (Verticality) and 11 (Eigentropy) to highlight probable locations of stems using those two metrics and the result will be a point cloud called "Stems". We use 50 closest points in ptm.knn which may be overkill and may slow down your processing depending on your machine.

all_metrics = fastPointMetrics.available() my_metrics = all_metrics[c(16, 11)] Stems = fastPointMetrics(thin, ptm.knn(50), my_metrics)

After that, we filter by values of interest. (high verticality and low eigentropy) and create a tree map using the treeMap function.

tlsfilter <- filter_poi(Stems, Verticality > 80, Verticality < 95) tlsPlot(tlsfilter) tlsfilter <- filter_poi(tlsfilter, Eigentropy < .03) tlsPlot(tlsfilter)

map = treeMap(tlsfilter, map.hough(min_h = 2, max_h= 4, min_votes = 1), merge = 0)

From there we plot the point cloud and classify tree regions, then stem points

x = plot(tls) add_treeMap(x, map, color='yellow', size=2)

classify tree regions tls = treePoints(tls, map, trp.crop()) treepoints <- add_treePoints(x, tls, size=4) plot(treepoints) add_treeIDs(x, tls, cex = 2, col='yellow')

classify stem points

tls = stemPoints(tls, stm.hough(h_step = 0.2, h_base = c(0.05, 0.25), min_votes = 1)) add_stemPoints(x, tls, color='red', size=8)