File size: 9,168 Bytes
4f6dd83 | 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 | """Level, centre and co-register a 2023/2025 crop pair.
Usage: python align.py raw2023.ply raw2025.ply out_dir [tag]
Writes out_dir/2023.ply, out_dir/2025.ply, out_dir/alignment.json"""
import sys, os, json, time
import numpy as np, pandas as pd
from scipy.spatial import cKDTree
def header_len(path):
n = 0
with open(path, 'rb') as f:
for line in f:
n += 1
if line.strip() == b'end_header': return n
def load(p):
return pd.read_csv(p, sep=r'\s+', header=None, skiprows=header_len(p), usecols=[0,1,2],
dtype=np.float64, engine='c').to_numpy()
def write_ply(path, pts, comments=()):
with open(path, 'w', newline='\n') as f:
f.write('ply\nformat ascii 1.0\n')
for c in comments: f.write(f'comment {c}\n')
f.write(f'element vertex {len(pts)}\nproperty float x\nproperty float y\nproperty float z\n'
'property uchar red\nproperty uchar green\nproperty uchar blue\nend_header\n')
np.savetxt(f, pts, fmt='%.4f %.4f %.4f 0 0 0')
def ground_points(P, cell=1.0):
"""lowest point per 1 m XY cell -> candidate ground samples"""
ij = np.floor(P[:, :2] / cell).astype(np.int64)
key = (ij[:, 0] - ij[:, 0].min()) * 100000 + (ij[:, 1] - ij[:, 1].min())
order = np.lexsort((P[:, 2], key))
key_s = key[order]
first = np.r_[True, key_s[1:] != key_s[:-1]]
return P[order[first]]
def ransac_plane(G, iters=400, thr=0.15, rng=np.random.default_rng(0)):
best_n, best_d, best_inl = None, None, -1
for _ in range(iters):
s = G[rng.choice(len(G), 3, replace=False)]
n = np.cross(s[1]-s[0], s[2]-s[0]); nn = np.linalg.norm(n)
if nn < 1e-9: continue
n /= nn
if n[2] < 0: n = -n
if n[2] < 0.7: continue # reject near-vertical planes (walls)
d = -n @ s[0]
dist = np.abs(G @ n + d)
inl = (dist < thr).sum()
if inl > best_inl: best_n, best_d, best_inl = n, d, inl
# refine with least squares on inliers
m = np.abs(G @ best_n + best_d) < thr
c = G[m].mean(0)
_, _, vt = np.linalg.svd(G[m] - c)
n = vt[2]; n = n if n[2] > 0 else -n
return n, -n @ c, best_inl / len(G)
def rot_to_z(n):
"""rotation matrix taking unit vector n to +z"""
z = np.array([0, 0, 1.0]); v = np.cross(n, z); s = np.linalg.norm(v); c = n @ z
if s < 1e-12: return np.eye(3)
vx = np.array([[0, -v[2], v[1]], [v[2], 0, -v[0]], [-v[1], v[0], 0]])
return np.eye(3) + vx + vx @ vx * ((1 - c) / s**2)
def kabsch(A, B):
ca, cb = A.mean(0), B.mean(0)
U, S, Vt = np.linalg.svd((A-ca).T @ (B-cb))
D = np.diag([1, 1, np.sign(np.linalg.det(Vt.T @ U.T))])
R = Vt.T @ D @ U.T
return R, cb - R @ ca
def icp(A, B, iters=60, trim=0.7, max_d=3.0, seed=0):
"""trimmed point-to-point ICP aligning A onto B; returns R,t (applied as A@R.T+t) and final median NN dist"""
rng = np.random.default_rng(seed)
sa = A[rng.choice(len(A), min(250000, len(A)), replace=False)]
sb = B[rng.choice(len(B), min(400000, len(B)), replace=False)]
tree = cKDTree(sb)
R, t = np.eye(3), np.zeros(3)
prev = np.inf
for it in range(iters):
cur = sa @ R.T + t
d, idx = tree.query(cur, workers=-1)
m = d < max_d
k = int(trim * m.sum())
sel = np.argsort(np.where(m, d, np.inf))[:k]
Ri, ti = kabsch(cur[sel], sb[idx[sel]])
R, t = Ri @ R, Ri @ t + ti
err = np.median(d[sel])
if abs(prev - err) < 1e-5: break
prev = err
cur = sa @ R.T + t
d, _ = tree.query(cur, workers=-1)
return R, t, float(np.median(d)), float(np.percentile(d, 90)), it + 1
def dsm(P, cell, x0, y0, nx, ny, zmax=15.0):
ix = np.floor((P[:, 0] - x0) / cell).astype(int); iy = np.floor((P[:, 1] - y0) / cell).astype(int)
m = (ix >= 0) & (ix < nx) & (iy >= 0) & (iy < ny) & (P[:, 2] > -1) & (P[:, 2] < zmax)
img = np.full((ny, nx), -np.inf)
np.maximum.at(img, (iy[m], ix[m]), P[m, 2])
occ = np.isfinite(img); img[~occ] = 0
return img, occ
def coarse_xy(A, B, cell=0.5, search=25.0):
"""XY shift (dx,dy) to apply to A so it best overlays B: cross-correlation of clipped DSMs"""
from scipy.signal import fftconvolve
lo = np.minimum(A[:, :2].min(0), B[:, :2].min(0)) - search; hi = np.maximum(A[:, :2].max(0), B[:, :2].max(0)) + search
nx, ny = int(np.ceil((hi[0]-lo[0]) / cell)), int(np.ceil((hi[1]-lo[1]) / cell))
ia, oa = dsm(A, cell, lo[0], lo[1], nx, ny); ib, ob = dsm(B, cell, lo[0], lo[1], nx, ny)
ia = (ia - ia[oa].mean()) * oa; ib = (ib - ib[ob].mean()) * ob
cc = fftconvolve(ib, ia[::-1, ::-1], mode='same')
ov = fftconvolve(ob.astype(float), oa[::-1, ::-1].astype(float), mode='same')
cc = cc / np.maximum(ov, 0.05 * ov.max())
cy, cx = np.array(cc.shape) // 2
r = int(search / cell)
win = cc[cy-r:cy+r+1, cx-r:cx+r+1]
py, px = np.unravel_index(np.argmax(win), win.shape)
return (px - r) * cell, (py - r) * cell
def level(P):
G = ground_points(P)
n, d, frac = ransac_plane(G)
R = rot_to_z(n)
p0 = np.array([0.0, 0.0, -d / n[2]]) # plane point under the XY origin
Q = (P - p0) @ R.T
tilt = np.degrees(np.arccos(np.clip(n[2], -1, 1)))
return Q, R, tilt, 0.0, frac
def main(f23, f25, out, tag='', pre=None, clip=False):
"""pre=(yaw_deg, [tx, ty]): 2D rigid transform p_map = Rot(yaw).p_loop + t linking the loop frame to the map frame;
the 2025 crop is first brought into the 2023 map frame with its inverse. clip=True keeps only 2025 points inside the
XY bounding box of the aligned 2023 crop, so both clouds cover the same footprint."""
t0 = time.time()
A = load(f23); B = load(f25)
if pre is not None:
yaw, tt = pre; c, s = np.cos(np.radians(yaw)), np.sin(np.radians(yaw))
Rz = np.array([[c, -s], [s, c]])
B[:, :2] = (B[:, :2] - np.asarray(tt)) @ Rz # inverse rotation: (p - t) @ R == R^T (p - t)
ctr = (np.r_[A[:, :2].min(0), 0] + np.r_[A[:, :2].max(0), 0]) / 2 # XY centre of the crop box (2023 box)
A = A - ctr; B = B - ctr
Bl, RB, tiltB, gzB, fracB = level(B); Bl[:, 2] -= gzB
Al, RA, tiltA, gzA, fracA = level(A); Al[:, 2] -= gzA
# centre XY of the levelled clouds on the 2023 crop
cB = np.r_[(Al[:, :2].min(0) + Al[:, :2].max(0)) / 2, 0]
Bl -= cB; Al -= cB
dx, dy = coarse_xy(Al, Bl)
Al = Al + np.array([dx, dy, 0.0])
R, t = np.eye(3), np.zeros(3); nit = 0
for md in (4.0, 2.0, 1.0, 0.5):
Ri, ti, med, p90, ni = icp(Al @ R.T + t, Bl, iters=30, max_d=md)
R, t = Ri @ R, Ri @ t + ti; nit += ni
t = t + R @ np.array([dx, dy, 0.0]) # fold the coarse shift into t (Al was shifted before)
Al = Al - np.array([dx, dy, 0.0])
Af = Al @ R.T + t
yaw = np.degrees(np.arctan2(R[1, 0], R[0, 0]))
rot = np.degrees(np.arccos(np.clip((np.trace(R) - 1) / 2, -1, 1)))
n_b_raw = len(B)
if clip:
lo, hi = Af[:, :2].min(0), Af[:, :2].max(0)
keep = (Bl[:, 0] >= lo[0]) & (Bl[:, 0] <= hi[0]) & (Bl[:, 1] >= lo[1]) & (Bl[:, 1] <= hi[1])
Bl = Bl[keep]
# reverse check
d_ba, _ = cKDTree(Af).query(Bl[np.random.default_rng(1).choice(len(Bl), min(200000, len(Bl)), replace=False)], workers=-1)
os.makedirs(out, exist_ok=True)
cm = [f'Subiaco change-detection pair {tag}'.strip(), 'Generated 2026-09-27 by align.py (level by RANSAC ground plane, XY-centre, trimmed ICP 2023->2025)']
write_ply(os.path.join(out, '2023.ply'), Af, cm + [f'source {os.path.basename(f23)}'])
write_ply(os.path.join(out, '2025.ply'), Bl, cm + [f'source {os.path.basename(f25)}'])
info = {'tag': tag, 'src2023': os.path.basename(f23), 'src2025': os.path.basename(f25),
'n2023': int(len(A)), 'n2025': int(len(Bl)), 'n2025_raw_crop': int(n_b_raw), 'clipped_to_2023_footprint': bool(clip),
'pre_transform_loop_to_map': None if pre is None else {'yaw_deg': float(pre[0]), 't': [float(v) for v in pre[1]]},
'xy_centre_raw': ctr[:2].round(3).tolist(),
'tilt2023_deg': round(float(tiltA), 3), 'tilt2025_deg': round(float(tiltB), 3),
'ground_inlier_frac_2023': round(float(fracA), 3), 'ground_inlier_frac_2025': round(float(fracB), 3),
'icp_rot_deg': round(float(rot), 3), 'icp_yaw_deg': round(float(yaw), 3), 'icp_t': t.round(3).tolist(), 'icp_iters': nit, 'coarse_xy_shift': [round(float(dx),2), round(float(dy),2)],
'nn_median_2023to2025_m': round(med, 3), 'nn_p90_2023to2025_m': round(p90, 3),
'nn_median_2025to2023_m': round(float(np.median(d_ba)), 3), 'nn_p90_2025to2023_m': round(float(np.percentile(d_ba, 90)), 3),
'bbox2023': [Af.min(0).round(2).tolist(), Af.max(0).round(2).tolist()],
'bbox2025': [Bl.min(0).round(2).tolist(), Bl.max(0).round(2).tolist()], 'seconds': round(time.time() - t0, 1)}
json.dump(info, open(os.path.join(out, 'alignment.json'), 'w'), indent=1)
print(json.dumps(info), flush=True)
return info
if __name__ == '__main__':
main(sys.argv[1], sys.argv[2], sys.argv[3], sys.argv[4] if len(sys.argv) > 4 else '')
|