terrio 0.5.0

Pure-Rust geospatial raster, vector and point cloud data I/O: read, write, convert and reproject GIS formats without C dependencies
Documentation 
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
260
261
262
263
264
265
266
267
268
269
270
271
272
273
274
275
276
277
278
279
280
281
282
283
284
285
286
287
288
289
290

//! Point clouds: PDAL's neighbourhood filters (dbscan, covariance features, eigenvalues,
//! approximate coplanarity, rank, local outlier factors, nearest-neighbour distances, radial
//! density, reciprocity, miniball, z smoothing), ordering and statistics filters (head, tail,
//! sort, locate, IQR, MAD), skewness balancing, and the filters that read rasters and vector
//! layers (dem, colorization, overlay).
//!
//! The synthetic tests run everywhere. The real-data tests read `autzen_trim.las` of
//! `scripts/fetch_test_data.sh` from `TERRIO_TEST_DATA` (they pass without checking anything when
//! it is not set); their expected values come from PDAL 2.9, and
//! `scripts/compare_pointclouds.py` compares every point with PDAL's.

use std::path::PathBuf;
use terrio::crs::SpatialRef;
use terrio::pointcloud::filters::{
    self, BandDimension, CovarianceFeaturesOptions, DbscanOptions, DimRange, EigenvaluesOptions, FeatureMode, Neighbourhood,
    NnDistanceMode, OverlayOptions,
};
use terrio::pointcloud::{self, Dimension, PointCloud, Values};
use terrio::{FieldDef, FieldType, FieldValue, Geometry, Layer, Raster, RasterConfig, RasterDataset};

fn data(name: &str) -> Option<String> {
    let dir = PathBuf::from(std::env::var_os("TERRIO_TEST_DATA")?);
    let p = dir.join("pointcloud").join(name);
    assert!(p.exists(), "{} is missing from TERRIO_TEST_DATA", p.display());
    Some(p.to_string_lossy().into_owned())
}

/// Points from coordinates, with intensities 0, 1, 2...
fn points(xyz: &[[f64; 3]]) -> PointCloud {
    let col = |k: usize| Values::F64(xyz.iter().map(|p| p[k]).collect());
    pointcloud::from_dimensions(
        vec![
            Dimension::new("x", col(0)),
            Dimension::new("y", col(1)),
            Dimension::new("z", col(2)),
            Dimension::new("intensity", Values::U16((0..xyz.len() as u16).collect())),
            Dimension::new("classification", Values::U8(vec![1; xyz.len()])),
        ],
        SpatialRef::default(),
    )
}

/// A square grid of `n × n` points one unit apart, on the plane z = 0.1 x + 0.2 y.
fn plane(n: usize) -> Vec<[f64; 3]> {
    (0..n * n)
        .map(|i| {
            let (x, y) = ((i % n) as f64, (i / n) as f64);
            [x, y, 0.1 * x + 0.2 * y]
        })
        .collect()
}

fn col(c: &PointCloud, name: &str) -> Vec<f64> {
    c.dim(name).unwrap_or_else(|| panic!("no {name}")).to_f64()
}

#[test]
fn clusters_by_density() {
    // Two dense rows of points and a lone point between them.
    let mut xyz: Vec<[f64; 3]> = (0..10).map(|i| [i as f64 * 0.5, 0.0, 0.0]).collect();
    xyz.push([20.0, 0.0, 0.0]);
    xyz.extend((0..10).map(|i| [40.0 + i as f64 * 0.5, 0.0, 0.0]));
    let c = filters::dbscan(&points(&xyz), &DbscanOptions { min_points: 3, ..Default::default() }).unwrap();
    let ids = c.dim("cluster_id").unwrap();
    assert!(matches!(ids.values, Values::I64(_)));
    assert_eq!(ids.to_f64(), [vec![0.0; 10], vec![-1.0], vec![1.0; 10]].concat());
    // In x alone, as in 3D here; too few points around each: all noise.
    let o = DbscanOptions { min_points: 30, dimensions: vec!["X".into()], ..Default::default() };
    assert!(col(&filters::dbscan(&points(&xyz), &o).unwrap(), "cluster_id").iter().all(|v| *v == -1.0));
    let o = DbscanOptions { dimensions: vec!["Nope".into()], ..Default::default() };
    assert!(filters::dbscan(&points(&xyz), &o).is_err());
}

#[test]
fn computes_covariance_features() {
    // A tilted plane: two equal eigenvalues and a third of 0, the normal off the vertical.
    let c = points(&plane(20));
    let o = CovarianceFeaturesOptions { feature_set: vec!["all".into(), "Density".into()], ..Default::default() };
    assert!(filters::covariance_features(&c, &o).is_err(), "Density needs OptimalKNN and OptimalRadius");
    let o = CovarianceFeaturesOptions {
        feature_set: vec!["Linearity".into(), "planarity".into(), " SCATTERING ".into(), "DemantkeVerticality".into(), "none".into()],
        mode: FeatureMode::Raw,
        ..Default::default()
    };
    let f = filters::covariance_features(&c, &o).unwrap();
    assert!(f.dim("verticality").is_none() && f.dim("none").is_none());
    let interior = |i: usize| (3..17).contains(&(i % 20)) && (3..17).contains(&(i / 20));
    let (scattering, demantke) = (col(&f, "scattering"), col(&f, "demantke_verticality"));
    let normal_z = 1.0 / (1.0f64 + 0.01 + 0.04).sqrt();
    for i in (0..400).filter(|&i| interior(i)) {
        assert!(scattering[i].abs() < 1e-9, "point {i}: {}", scattering[i]);
        // The offsets from the mean are rounded to floats, as PDAL rounds them.
        assert!((demantke[i] - (1.0 - normal_z)).abs() < 1e-7, "point {i}: {}", demantke[i]);
    }
    // A line: linearity 1.
    let line: Vec<[f64; 3]> = (0..50).map(|i| [i as f64, 2.0 * i as f64, 0.5 * i as f64]).collect();
    let f = filters::covariance_features(&points(&line), &CovarianceFeaturesOptions::default()).unwrap();
    assert!(col(&f, "linearity").iter().all(|v| (v - 1.0).abs() < 1e-6));
    assert!(FeatureMode::parse("SQRT").is_ok() && FeatureMode::parse("cube").is_err());
}

#[test]
fn computes_eigenvalues_and_ranks() {
    let c = points(&plane(15));
    let o = EigenvaluesOptions { normalize: true, ..Default::default() };
    let e = filters::eigenvalues(&c, &o).unwrap();
    let sums: Vec<f64> = (0..c.len()).map(|i| (0..3).map(|k| e.dim(&format!("eigenvalue{k}")).unwrap().get(i)).sum()).collect();
    assert!(sums.iter().all(|s| (s - 1.0).abs() < 1e-12));
    assert!(col(&e, "eigenvalue0").iter().all(|v| v.abs() < 1e-9));
    // With a radius: the neighbourhoods with fewer than min_k points are skipped (0).
    let o = EigenvaluesOptions { neighbours: Neighbourhood { knn: 8, stride: 1, radius: Some(0.5), min_k: 3 }, normalize: false };
    assert!(col(&filters::eigenvalues(&c, &o).unwrap(), "eigenvalue2").iter().all(|v| *v == 0.0));
    // A plane is of rank 2, approximately coplanar.
    assert!(col(&filters::estimate_rank(&c, 8, 0.01).unwrap(), "rank").iter().all(|v| *v == 2.0));
    let coplanar = col(&filters::approximate_coplanar(&c, 8, 25.0, 6.0).unwrap(), "coplanar");
    assert!(coplanar.iter().filter(|v| **v == 1.0).count() > 150);
}

#[test]
fn measures_nearest_neighbours() {
    // A grid one unit apart, and a point far from it.
    let mut xyz: Vec<[f64; 3]> = (0..100).map(|i| [(i % 10) as f64, (i / 10) as f64, 0.0]).collect();
    xyz.push([100.0, 100.0, 0.0]);
    let c = points(&xyz);
    let lof = filters::lof(&c, 5).unwrap();
    let factors = col(&lof, "local_outlier_factor");
    let max = factors.iter().copied().fold(f64::MIN, f64::max);
    assert_eq!(factors[100], max, "the lone point is the most outlying");
    assert!(lof.dim("nndistance").is_some() && lof.dim("local_reachability_distance").is_some());
    // The 4 nearest neighbours of an interior grid point are 1 away.
    let d = col(&filters::nndistance(&c, 4, NnDistanceMode::Kth).unwrap(), "nndistance");
    assert_eq!(d[55], 1.0);
    let d = col(&filters::nndistance(&c, 8, NnDistanceMode::Avg).unwrap(), "nndistance");
    assert!((d[55] - (4.0 + 4.0 * 2f64.sqrt()) / 8.0).abs() < 1e-12);
    // Within 1.1: the point itself and its 4 neighbours.
    let density = col(&filters::radial_density(&c, 1.1).unwrap(), "radial_density");
    let pdal_pi = 314_159.0 / 100_000.0;
    assert!((density[55] - 5.0 / (4.0 / 3.0 * pdal_pi * 1.1f64.powi(3))).abs() < 1e-12);
    assert_eq!(col(&filters::reciprocity(&c, 4).unwrap(), "reciprocity")[100], 100.0);
    let m = col(&filters::miniball(&c, 8).unwrap(), "miniball");
    assert!(m[55].abs() < 1e-12, "a grid point is at the centre of its 8 neighbours");
    assert!(NnDistanceMode::parse("AVG").is_ok() && NnDistanceMode::parse("max").is_err());
}

#[test]
fn smooths_elevations() {
    // Three points at each place, a spike in the middle.
    let xyz = vec![[0.0, 0.0, 1.0], [0.5, 0.0, 2.0], [0.0, 0.5, 3.0], [0.2, 0.2, 100.0], [10.0, 10.0, 7.0]];
    let c = filters::zsmooth(&points(&xyz), "Smoothed", 1.0, 50.0).unwrap();
    let s = col(&c, "smoothed");
    assert_eq!(s[3], 2.0, "the median of 1, 2 and 3");
    assert_eq!(s[4], 7.0, "a point alone keeps its z");
    let low = col(&filters::zsmooth(&points(&xyz), "Smoothed", 1.0, 0.0).unwrap(), "smoothed");
    assert_eq!(low[3], 1.0);
    assert!(filters::zsmooth(&points(&xyz), "Z", 1.0, 50.0).is_err());
}

#[test]
fn balances_skewness() {
    // Ground near 0, then trees: the points come out sorted by z, the low ones ground.
    let mut xyz: Vec<[f64; 3]> = (0..200).map(|i| [i as f64, 0.0, ((i * 37) % 11) as f64 * 0.01]).collect();
    xyz.extend((0..40).map(|i| [i as f64, 1.0, 10.0 + i as f64]));
    let c = filters::skewness_balancing(&points(&xyz), 2, 1, false).unwrap();
    let z = col(&c, "z");
    assert!(z.windows(2).all(|w| w[0] <= w[1]));
    // The ground is the lowest points, the trees among the others.
    let classes = col(&c, "classification");
    let ground = classes.iter().take_while(|c| **c == 2.0).count();
    assert!(ground > 0 && ground < 200 && classes[ground..].iter().all(|c| *c == 1.0), "{ground}");
    assert!(filters::skewness_balancing(&points(&xyz), 2, 2, false).is_err());
}

#[test]
fn selects_and_orders_points() {
    let xyz: Vec<[f64; 3]> = (0..20).map(|i| [i as f64, (i % 3) as f64, ((i * 7) % 20) as f64]).collect();
    let c = points(&xyz);
    assert_eq!(col(&filters::head(&c, 3, false), "x"), [0.0, 1.0, 2.0]);
    assert_eq!(filters::head(&c, 3, true).len(), 17);
    assert_eq!(col(&filters::tail(&c, 2, false), "x"), [18.0, 19.0]);
    assert_eq!(filters::tail(&c, 50, false).len(), 20);
    assert_eq!(filters::tail(&c, 50, true).len(), 0);
    let s = filters::sort(&c, &["Z".into()], true).unwrap();
    assert!(col(&s, "z").windows(2).all(|w| w[0] >= w[1]));
    // By y, then x: the last dimension orders, the first breaks ties.
    let s = filters::sort(&c, &["X".into(), "Y".into()], false).unwrap();
    assert_eq!(col(&s, "x")[..7], [0.0, 3.0, 6.0, 9.0, 12.0, 15.0, 18.0]);
    assert_eq!(col(&filters::locate(&c, "Z", false).unwrap(), "z"), [19.0]);
    assert_eq!(col(&filters::locate(&c, "z", true).unwrap(), "x"), [0.0]);
    // An outlier beyond the fences.
    let mut xyz = xyz;
    xyz.push([0.0, 0.0, 1000.0]);
    let c = points(&xyz);
    assert_eq!(filters::iqr(&c, "Z", 1.5).unwrap().len(), 20);
    assert_eq!(filters::mad(&c, "Z", 2.0, 1.4862).unwrap().len(), 20);
    assert!(filters::sort(&c, &[], false).is_err() && filters::locate(&c, "nope", true).is_err());
}

/// A 10 × 10 raster of 1-unit cells from (0, 0), two bands: x + 10 y, and 100 − that.
fn raster() -> RasterDataset {
    let cfg = RasterConfig { cols: 10, rows: 10, bands: 2, x_min: 0.0, y_min: 0.0, cell_size: 1.0, ..Default::default() };
    let mut data = Vec::new();
    for band in 0..2 {
        for row in 0..10 {
            for c in 0..10 {
                // Rows from the top: y of the row's centre is 9.5 - row.
                let v = c as f64 + 10.0 * (9 - row) as f64;
                data.push(if band == 0 { v } else { 100.0 - v });
            }
        }
    }
    RasterDataset::new(Raster::from_data(cfg, data).unwrap(), true)
}

#[test]
fn keeps_points_near_a_raster_surface() {
    // Points at the cell centres, z the cell's value plus 0, 1, 2 or 3; one outside.
    let mut xyz: Vec<[f64; 3]> = (0..100).map(|i| [(i % 10) as f64 + 0.5, (i / 10) as f64 + 0.5, i as f64 + (i % 4) as f64]).collect();
    xyz.push([20.0, 20.0, 0.0]);
    let c = points(&xyz);
    let kept = filters::dem(&c, &raster(), &DimRange::parse("Z[0:1.5]").unwrap(), 1).unwrap();
    assert_eq!(kept.len(), 50, "z up to 1.5 above the surface");
    assert!(filters::dem(&c, &raster(), &DimRange::parse("Z[0:1]").unwrap(), 3).is_err());
}

#[test]
fn colorizes_points_from_raster_bands() {
    let xyz = vec![[0.5, 9.5, 0.0], [3.2, 0.1, 0.0], [10.5, 5.0, 0.0]];
    let dims = BandDimension::parse_list("Red:1:2, Shade:2").unwrap();
    let c = filters::colorization(&points(&xyz), &raster(), &dims).unwrap();
    // Doubles, as PDAL makes them; the point outside keeps 0.
    assert!(matches!(c.dim("red").unwrap().values, Values::F64(_)));
    assert_eq!(col(&c, "red"), [180.0, 6.0, 0.0]);
    assert_eq!(col(&c, "shade"), [10.0, 97.0, 0.0]);
    let rgb = BandDimension::rgb();
    assert!(filters::colorization(&points(&xyz), &raster(), &rgb).is_err(), "no third band");
}

#[test]
fn overlays_polygon_attributes() {
    let mut layer = Layer::new("zones");
    layer.add_field(FieldDef::new("name", FieldType::Text));
    layer.add_field(FieldDef::new("zone", FieldType::Integer));
    let square = |x0: f64, holes: Vec<Vec<wbvector::Coord>>| {
        let ring = [(x0, 0.0), (x0 + 10.0, 0.0), (x0 + 10.0, 10.0), (x0, 10.0), (x0, 0.0)];
        Geometry::polygon(ring.iter().map(|&(x, y)| wbvector::Coord::xy(x, y)).collect(), holes)
    };
    let hole: Vec<wbvector::Coord> =
        [(2.0, 2.0), (4.0, 2.0), (4.0, 4.0), (2.0, 4.0), (2.0, 2.0)].iter().map(|&(x, y)| wbvector::Coord::xy(x, y)).collect();
    for (k, g) in [square(0.0, vec![hole]), square(5.0, vec![])].into_iter().enumerate() {
        let attributes = vec![FieldValue::Text(format!("{}7", k + 1)), FieldValue::Integer(k as i64 + 3)];
        layer.push(terrio::Feature { fid: k as u64, geometry: Some(g), attributes });
    }
    let xyz = vec![[1.0, 1.0, 0.0], [3.0, 3.0, 0.0], [7.0, 5.0, 0.0], [12.0, 5.0, 0.0], [30.0, 5.0, 0.0], [0.0, 5.0, 0.0]];
    let c = points(&xyz);
    // By default the second attribute: the first polygon has the point, the hole does not.
    let o = OverlayOptions { dimension: "Classification".into(), column: None };
    assert_eq!(col(&filters::overlay(&c, &layer, &o).unwrap(), "classification"), [3.0, 1.0, 3.0, 4.0, 1.0, 3.0]);
    let o = OverlayOptions { dimension: "Intensity".into(), column: Some("NAME".into()) };
    assert_eq!(col(&filters::overlay(&c, &layer, &o).unwrap(), "intensity"), [17.0, 1.0, 17.0, 27.0, 4.0, 17.0]);
    assert!(filters::overlay(&c, &layer, &OverlayOptions { dimension: "Nope".into(), column: None }).is_err());
    assert!(filters::overlay(&c, &layer, &OverlayOptions { dimension: "Intensity".into(), column: Some("x".into()) }).is_err());
}

fn autzen() -> Option<PointCloud> {
    Some(pointcloud::read_points(&data("autzen_trim.las")?, &Default::default()).unwrap())
}

#[test]
fn neighbourhood_filters_as_pdal_does_on_autzen() {
    // PDAL 2.9's filters on autzen_trim.las (110 000 points, in feet).
    let Some(c) = autzen() else { return };
    let sum = |c: &PointCloud, name: &str| col(c, name).iter().sum::<f64>();
    let o = DbscanOptions { eps: 3.0, min_points: 10, ..Default::default() };
    let ids = col(&filters::dbscan(&c, &o).unwrap(), "cluster_id");
    assert_eq!((ids.iter().copied().fold(f64::MIN, f64::max), ids.iter().filter(|v| **v == -1.0).count()), (359.0, 95_169));
    let e = filters::eigenvalues(&c, &EigenvaluesOptions::default()).unwrap();
    assert_eq!(e.dim("eigenvalue2").unwrap().get(0), 16.185_960_628_220_393);
    let f = filters::covariance_features(&c, &CovarianceFeaturesOptions::default()).unwrap();
    assert_eq!(f.dim("linearity").unwrap().get(0), 0.677_748_811_127_886_5);
    assert_eq!(f.dim("verticality").unwrap().get(0), 0.032_007_386_653_545_196);
    let lof = filters::lof(&c, 10).unwrap();
    assert_eq!(lof.dim("local_outlier_factor").unwrap().get(0), 1.031_589_756_323_879_4);
    assert_eq!(sum(&filters::reciprocity(&c, 8).unwrap(), "reciprocity"), 1_061_100.0);
    assert_eq!(col(&filters::estimate_rank(&c, 8, 0.01).unwrap(), "rank").iter().filter(|r| **r == 3.0).count(), 37_754);
    assert_eq!(col(&filters::approximate_coplanar(&c, 8, 25.0, 6.0).unwrap(), "coplanar").iter().filter(|r| **r == 1.0).count(), 75_369);
    assert_eq!(filters::iqr(&c, "Z", 1.5).unwrap().len(), 82_454);
    assert_eq!(filters::mad(&c, "Z", 2.0, 1.4862).unwrap().len(), 80_067);
    let ground = filters::skewness_balancing(&c, 2, 1, false).unwrap();
    assert_eq!(col(&ground, "classification").iter().filter(|v| **v == 2.0).count(), 102_234);
}