有時候需要判斷某些點是否在點雲的凸包內，例如3d object detection 中找到了bbox，判斷場景中的點雲哪些在這個bbox中，這裡以2d點為例：
import numpy as np
import matplotlib.pyplot as plt
from matplotlib.collections import polycollection, linecollection
from scipy.spatial import delaunay
lidar_file =
"/home/seivl/pcdet/data/kitti/training/velodyne/000000.bin"
pts = np.fromfile(lidar_file, dtype=np.float32)
.reshape(-1
,4)print
(pts)
print
(pts.shape)
defin_hull
(p, hull)
:"""
test if points in `p` are in `hull`
`p` should be a `nxk` coordinates of `n` points in `k` dimensions
`hull` is either a scipy.spatial.delaunay object or the `mxk` array of the
coordinates of `m` points in `k`dimensions for which delaunay triangulation
will be computed
"""from scipy.spatial import delaunay
ifnotisinstance
(hull,delaunay)
: hull = delaunay(hull)
return hull.find_******x(p)
>=
0, hull.find_******x(p)
<
0def
plot_in_hull
(p, hull)
:"""
plot relative to `in_hull` for 2d data
"""ifnot
isinstance
(hull,delaunay)
: hull = delaunay(hull)
# plot triangulation
poly = polycollection(hull.points[hull.vertices]
, facecolors=
'w', edgecolors=
'b')
plt.clf(
) plt.title(
'in hull'
) plt.gca(
).add_collection(poly)
# plt.plot(hull.points[:,0], hull.points[:,1], 'o', hold=1)
plt.plot(hull.points[:,
0], hull.points[:,
1],'o'
)# plot the convex hull
edges =
set(
) edge_points =
defadd_edge
(i, j)
:"""add a line between the i-th and j-th points, if not in the list already"""
if(i, j)
in edges or
(j, i)
in edges:
# already added
return
edges.add(
(i, j)
)[i, j]])
for ia, ib in hull.convex_hull:
add_edge(ia, ib)
lines = linecollection(edge_points, color=
'g')
plt.gca(
).add_collection(lines)
plt.show(
)# plot tested points `p` - black are inside hull, red outside
inside, out = in_hull(p,hull)
plt.plot(p[inside,0]
,p[inside,1]
,'.k'
) plt.plot(p[out,0]
,p[out,1]
,'.r'
)# plt.plot(p[-inside,0],p[-inside,1],'.r')
plt.show(
)if __name__ ==
'__main__'
: p1 = pts[
2000,0
:2] p2 = pts[
12782,0
:2] p3 = pts[
48772,0
:2] p4 = pts[
38217,0
:2] p_in = np.vstack(
(p1, p2, p3, p4)
) plot_in_hull(pts[:,
0:2]
, p_in)
# 方法 2
3D目標檢測之AM3D
accurate monocular object detection via color embedded 3d reconstructionfor autonomous driving 參考戳這裡 先用2d目標檢測網路和單目深度估計網路分別檢測2d box和深度資訊，然後將深度資訊轉化為三維點雲...
單目3D目標檢測調研
現有的單目3d目標檢測方案主要方案主要分為兩類，分別為基於的方法和基於偽雷達點雲的方法。基於的方法一般通過2d 3d之間的幾何約束來學習，包括目標形狀資訊，地面資訊，以及關鍵點等等，通過這些資訊在損失函式中約束3d檢測學習得更好。基於偽雷達點雲的方法首先從2d中估計出單目深度資訊來，然後利用單目深度...
3D點雲目標的識別與抓取
這是2016年乙個點雲目標識別研究的簡單描述。點雲目標識別，顧名思義，需要有標準的目標點雲或者標準的點雲特徵描述向量 對實時採集的點雲資料，在裡面尋找與目標點雲相似度最高的點雲塊。圖2.1 點雲在xy平面的投影圖像 圖2.2 圓環點雲採集影象 常用點雲濾波有帶通濾波 統計濾波 半徑濾波 去除背景平面...