def test_downsample_random():
pc = pcl.PointCloud(10)
a = np.asarray(pc)
a[:] = np.random.randn(*a.shape)
assert_raises(ValueError, utils.downsample_random, pc, 0)
assert_raises(ValueError, utils.downsample_random, pc, 2)
assert_equal(len(utils.downsample_random(pc, 0.39)), 4)
def test_downsample_random():
pc = pcl.PointCloud(10)
a = np.asarray(pc)
a[:] = np.random.randn(*a.shape)
assert_raises(ValueError, utils.downsample_random, pc, 0)
assert_raises(ValueError, utils.downsample_random, pc, 2)
assert_equal(len(utils.downsample_random(pc, .39)), 4)
def boundary_of_center_object(pc,
downsample=None,
angle_threshold=0.1,
search_radius=0.1,
normal_search_radius=0.1):
"""
Find the boundary of the main object.
First applies dbscan to find the main object,
then estimates its footprint by taking the pointcloud boundary.
Resulting pointcloud has the same SRS and offset as the input.
Arguments:
pointcloud : pcl.PointCloud
downsample : If given, reduce the pointcloud to given percentage
values should be in [0,1]
angle_threshold : float defaults to 0.1
search_radius : float defaults to 0.1
normal_search_radius : float defaults to 0.1
Returns:
boundary : pcl.PointCloud
"""
if downsample is not None:
log(' - Random downsampling factor:', downsample)
pc = utils.downsample_random(pc, downsample)
else:
log(' - Not downsampling')
# Main noise supression step
# Find largest clusters, accounting for at least 70% of the pointcloud.
# Presumably, this is the main object.
log(' - Starting dbscan on downsampled pointcloud')
mainobject = get_largest_dbscan_clusters(pc, 0.7, .075, 250)
save(mainobject, 'mainobject.las')
boundary = estimate_boundaries(mainobject,
angle_threshold=angle_threshold,
search_radius=search_radius,
normal_search_radius=normal_search_radius)
boundary = extract_mask(mainobject, boundary)
if len(boundary) == len(mainobject) or len(boundary) == 0:
log(' - Cannot find boundary')
return None
# project on the xy plane, take 2th percentile height
points = np.asarray(boundary)
points[:, 2] = np.percentile(points[:, 2], 2)
# Secondary noise supression step
# some cases have multiple objects close to eachother, and we need to
# filter some out. Assume the object is a single item; perform another
# dbscan to select the footprint of the main item
log(' - Starting dbscan on boundary')
mainboundary = get_largest_dbscan_clusters(boundary, 0.5, .1, 10)
# Evenly space out the points
mainboundary = utils.downsample_voxel(mainboundary, voxel_size=0.1)
utils.force_srs(mainboundary, same_as=pc)
return mainboundary
