Register fragments#

Once the fragments of the scene are created, the next step is to align them in a global space.

Input arguments#

This script runs with python run_system.py [config] --register. In [config], ["path_dataset"] should have subfolders fragments which stores fragments in .ply files and a pose graph in a .json file.

The main function runs make_posegraph_for_scene and optimize_posegraph_for_scene. The first function performs pairwise registration. The second function performs multiway registration.

Preprocess point cloud#

27def preprocess_point_cloud(pcd, config):
28    voxel_size = config["voxel_size"]
29    pcd_down = pcd.voxel_down_sample(voxel_size)
30    pcd_down.estimate_normals(
31        cv3d.geometry.KDTreeSearchParamHybrid(radius=voxel_size * 2.0,
32                                              max_nn=30))
33    pcd_fpfh = cv3d.pipelines.registration.compute_fpfh_feature(
34        pcd_down,
35        cv3d.geometry.KDTreeSearchParamHybrid(radius=voxel_size * 5.0,
36                                              max_nn=100))
37    return (pcd_down, pcd_fpfh)

This function downsamples a point cloud to make it sparser and regularly distributed. Normals and FPFH feature are precomputed. See Voxel downsampling, Vertex normal estimation, and Extract geometric feature for more details.

Compute initial registration#

70def compute_initial_registration(s, t, source_down, target_down, source_fpfh,
71                                 target_fpfh, path_dataset, config):
72
73    if t == s + 1:  # odometry case
74        print("Using RGBD odometry")
75        pose_graph_frag = cv3d.io.read_pose_graph(
76            join(path_dataset,
77                 config["template_fragment_posegraph_optimized"] % s))
78        n_nodes = len(pose_graph_frag.nodes)
79        transformation_init = np.linalg.inv(pose_graph_frag.nodes[n_nodes -
80                                                                  1].pose)
81        (transformation, information) = \
82                multiscale_icp(source_down, target_down,
83                [config["voxel_size"]], [50], config, transformation_init)
84    else:  # loop closure case
85        (success, transformation,
86         information) = register_point_cloud_fpfh(source_down, target_down,
87                                                  source_fpfh, target_fpfh,
88                                                  config)
89        if not success:
90            print("No reasonable solution. Skip this pair")
91            return (False, np.identity(4), np.zeros((6, 6)))
92    print(transformation)
93
94    if config["debug_mode"]:
95        draw_registration_result(source_down, target_down, transformation)
96    return (True, transformation, information)

This function computes a rough alignment between two fragments. If the fragments are neighboring fragments, the rough alignment is determined by an aggregating RGBD odometry obtained from Make fragments. Otherwise, register_point_cloud_fpfh is called to perform global registration. Note that global registration is less reliable according to [Choi2015].

Pairwise global registration#

40def register_point_cloud_fpfh(source, target, source_fpfh, target_fpfh, config):
41    cv3d.utility.set_verbosity_level(cv3d.utility.VerbosityLevel.Debug)
42    distance_threshold = config["voxel_size"] * 1.4
43    if config["global_registration"] == "fgr":
44        result = cv3d.pipelines.registration.registration_fgr_based_on_feature_matching(
45            source, target, source_fpfh, target_fpfh,
46            cv3d.pipelines.registration.FastGlobalRegistrationOption(
47                maximum_correspondence_distance=distance_threshold))
48    if config["global_registration"] == "ransac":
49        # Fallback to preset parameters that works better
50        result = cv3d.pipelines.registration.registration_ransac_based_on_feature_matching(
51            source, target, source_fpfh, target_fpfh, False, distance_threshold,
52            cv3d.pipelines.registration.TransformationEstimationPointToPoint(
53                False), 4, [
54                    cv3d.pipelines.registration.
55                    CorrespondenceCheckerBasedOnEdgeLength(0.9),
56                    cv3d.pipelines.registration.
57                    CorrespondenceCheckerBasedOnDistance(distance_threshold)
58                ],
59            cv3d.pipelines.registration.RANSACConvergenceCriteria(
60                1000000, 0.999))
61    if (result.transformation.trace() == 4.0):
62        return (False, np.identity(4), np.zeros((6, 6)))
63    information = cv3d.pipelines.registration.get_information_matrix_from_point_clouds(
64        source, target, distance_threshold, result.transformation)
65    if information[5, 5] / min(len(source.points), len(target.points)) < 0.3:
66        return (False, np.identity(4), np.zeros((6, 6)))
67    return (True, result.transformation, information)

This function uses RANSAC or Fast global registration for pairwise global registration.

Multiway registration#

 99def update_posegraph_for_scene(s, t, transformation, information, odometry,
100                               pose_graph):
101    if t == s + 1:  # odometry case
102        odometry = np.dot(transformation, odometry)
103        odometry_inv = np.linalg.inv(odometry)
104        pose_graph.nodes.append(
105            cv3d.pipelines.registration.PoseGraphNode(odometry_inv))
106        pose_graph.edges.append(
107            cv3d.pipelines.registration.PoseGraphEdge(s,
108                                                      t,
109                                                      transformation,
110                                                      information,
111                                                      uncertain=False))
112    else:  # loop closure case
113        pose_graph.edges.append(
114            cv3d.pipelines.registration.PoseGraphEdge(s,
115                                                      t,
116                                                      transformation,
117                                                      information,
118                                                      uncertain=True))
119    return (odometry, pose_graph)

This script uses the technique demonstrated in Multiway registration. The function update_posegraph_for_scene builds a pose graph for multiway registration of all fragments. Each graph node represents a fragment and its pose which transforms the geometry to the global space.

Once a pose graph is built, the function optimize_posegraph_for_scene is called for multiway registration.

46def optimize_posegraph_for_scene(path_dataset, config):
47    pose_graph_name = join(path_dataset, config["template_global_posegraph"])
48    pose_graph_optimized_name = join(
49        path_dataset, config["template_global_posegraph_optimized"])
50    run_posegraph_optimization(pose_graph_name, pose_graph_optimized_name,
51            max_correspondence_distance = config["voxel_size"] * 1.4,
52            preference_loop_closure = \
53            config["preference_loop_closure_registration"])

Main registration loop#

The function make_posegraph_for_scene below calls all the functions introduced above. The main workflow is: pairwise global registration -> multiway registration.

152def make_posegraph_for_scene(ply_file_names, config):
153    pose_graph = cv3d.pipelines.registration.PoseGraph()
154    odometry = np.identity(4)
155    pose_graph.nodes.append(cv3d.pipelines.registration.PoseGraphNode(odometry))
156
157    n_files = len(ply_file_names)
158    matching_results = {}
159    for s in range(n_files):
160        for t in range(s + 1, n_files):
161            matching_results[s * n_files + t] = matching_result(s, t)
162
163    if config["python_multi_threading"] is True:
164        os.environ['OMP_NUM_THREADS'] = '1'
165        max_workers = max(
166            1, min(multiprocessing.cpu_count() - 1, len(matching_results)))
167        mp_context = multiprocessing.get_context('spawn')
168        with mp_context.Pool(processes=max_workers) as pool:
169            args = [(ply_file_names, v.s, v.t, config)
170                    for k, v in matching_results.items()]
171            results = pool.starmap(register_point_cloud_pair, args)
172
173        for i, r in enumerate(matching_results):
174            matching_results[r].success = results[i][0]
175            matching_results[r].transformation = results[i][1]
176            matching_results[r].information = results[i][2]
177    else:
178        for r in matching_results:
179            (matching_results[r].success, matching_results[r].transformation,
180             matching_results[r].information) = \
181                register_point_cloud_pair(ply_file_names,
182                                          matching_results[r].s, matching_results[r].t, config)
183
184    for r in matching_results:
185        if matching_results[r].success:
186            (odometry, pose_graph) = update_posegraph_for_scene(
187                matching_results[r].s, matching_results[r].t,
188                matching_results[r].transformation,
189                matching_results[r].information, odometry, pose_graph)
190    cv3d.io.write_pose_graph(
191        join(config["path_dataset"], config["template_global_posegraph"]),
192        pose_graph)

Results#

The pose graph optimization outputs the following messages:

[GlobalOptimizationLM] Optimizing PoseGraph having 14 nodes and 42 edges.
Line process weight : 55.885667
[Initial     ] residual : 7.791139e+04, lambda : 1.205976e+00
[Iteration 00] residual : 6.094275e+02, valid edges : 22, time : 0.001 sec.
[Iteration 01] residual : 4.526879e+02, valid edges : 22, time : 0.000 sec.
[Iteration 02] residual : 4.515039e+02, valid edges : 22, time : 0.000 sec.
[Iteration 03] residual : 4.514832e+02, valid edges : 22, time : 0.000 sec.
[Iteration 04] residual : 4.514825e+02, valid edges : 22, time : 0.000 sec.
Current_residual - new_residual < 1.000000e-06 * current_residual
[GlobalOptimizationLM] total time : 0.003 sec.
[GlobalOptimizationLM] Optimizing PoseGraph having 14 nodes and 35 edges.
Line process weight : 60.762800
[Initial     ] residual : 6.336097e+01, lambda : 1.324043e+00
[Iteration 00] residual : 6.334147e+01, valid edges : 22, time : 0.000 sec.
[Iteration 01] residual : 6.334138e+01, valid edges : 22, time : 0.000 sec.
Current_residual - new_residual < 1.000000e-06 * current_residual
[GlobalOptimizationLM] total time : 0.001 sec.
CompensateReferencePoseGraphNode : reference : 0

There are 14 fragments and 52 valid matching pairs among the fragments. After 23 iterations, 11 edges are detected to be false positives. After they are pruned, pose graph optimization runs again to achieve tight alignment.