Method, apparatus, device and computer readable storage medium for reconstructing three-dimensional scene

1. A method for reconstructing a three-dimensional scene, comprising:

acquiring a point cloud data frame set for the three-dimensional scene, point cloud data frames in the point cloud data frame set respectively having a pose parameter;

determining a subset corresponding to a part of the three-dimensional scene from the point cloud data frame set, comprising:

determining, based on trajectories corresponding to the point cloud data frame set, a position of an intersection in the three-dimensional scene, comprising: performing dilation on a trajectory image comprising the trajectories to determine a line representing a road by dilating trajectories corresponding to the road; determining, based on the line, at least one corner point at the intersection; and determining the position of the intersection based on the at least one corner point at the intersection;

forming, based on the position of the intersection, a connection diagram connecting the intersection; and

determining, based on the connection diagram, a closed path as the part of the three-dimensional scene;

updating the subset by removing data points associated with a dynamic object from point cloud data frames in the subset;

adjusting a pose parameter of a point cloud data frame in the updated subset to obtain an adjusted subset, the adjusted subset including at least two point cloud data frames having matching overlapping parts; and

updating the point cloud data frame set using the adjusted subset.

2. The method according to claim 1 , wherein the removing data points comprises:

removing data points representing a ground from the point cloud data frames;

recognizing the dynamic object in the point cloud data frames;

removing data points representing the dynamic object from the point cloud data frames; and

adding the data points representing the ground into the point cloud data frames.

3. The method according to claim 2 , wherein the recognizing the dynamic object comprises:

clustering point cloud data in the point cloud data frames to determine a plurality of clusters in the point cloud data frames; and

determining a cluster matching an attribute of the dynamic object from the plurality of clusters.

4. The method according to claim 1 , wherein the adjusting a pose parameter comprises:

extracting data points representing a static object from point cloud data frames in the subset, to update the subset; and

adjusting a pose parameter of a point cloud data frame in the updated subset.

5. The method according to claim 1 , wherein the at least two point cloud data frames include a first point cloud data frame and a second point cloud data frame, and the adjusting a pose parameter comprises:

determining a characteristic distance between corresponding objects in the first point cloud data frame and the second point cloud data frame; and

adjusting the pose parameter based on the characteristic distance.

6. The method according to claim 1 , wherein the updating the point cloud data frame set comprises:

determining another part of the three-dimensional scene, the another part overlapping the part of the three-dimensional scene;

determining another subset corresponding to the another part of the three-dimensional scene from the point cloud data frame set; and

adjusting a pose parameter of a point cloud data frame in the subset and a pose parameter of a point cloud data frame in the another subset, to cause a third point cloud data frame in the subset to overlap a fourth point cloud data frame in the another subset.

7. The method according to claim 1 , where the updating the point cloud data frame set comprises:

determining a pose parameter of a fifth point cloud data frame in the part through a global navigation satellite system; and

updating the point cloud data frame set based on the pose parameter of the fifth point cloud data frame.

8. An apparatus for reconstructing a three-dimensional scene, comprising:

at least one processor; and

a memory storing instructions, the instructions when executed by the at least one processor, cause the at least one processor to perform operations, the operations comprising:

acquiring a point cloud data frame set for the three-dimensional scene, point cloud data frames in the point cloud data frame set respectively having a pose parameter;

determining a subset corresponding to a part of the three-dimensional scene from the point cloud data frame set, comprising:

determining, based on trajectories corresponding to the point cloud data frame set, a position of an intersection in the three-dimensional scene, comprising: performing dilation on a trajectory image comprising the trajectories to determine a line representing a road by dilating trajectories corresponding to the road; determining, based on the line, at least one corner point at the intersection; and determining the position of the intersection based on the at least one corner point at the intersection;

forming, based on the position of the intersection, a connection diagram connecting the intersection; and

determining, based on the connection diagram, a closed path as the part of the three-dimensional scene;

updating the subset by removing data points associated with a dynamic object from point cloud data frames in the subset;

adjusting a pose parameter of a point cloud data frame in the updated subset to obtain an adjusted subset, the adjusted subset including at least two point cloud data frames having matching overlapping parts; and

updating the point cloud data frame set using the adjusted subset.

9. The apparatus according to claim 8 , wherein the removing data points comprises:

removing data points representing a ground from the point cloud data frames;

recognizing the dynamic object in the point cloud data frames;

removing data points representing the dynamic object from the point cloud data frames; and

adding the data points representing the ground into the point cloud data frames,

wherein the recognizing the dynamic object comprises:

clustering point cloud data in the point cloud data frames to determine a plurality of clusters in the point cloud data frames; and

determining a cluster matching an attribute of the dynamic object from the plurality of clusters.

10. The apparatus according to claim 8 , wherein the adjusting a pose parameter comprises:

extracting data points representing a static object from point cloud data frames in the subset, to update the subset; and

adjusting a pose parameter of a point cloud data frame in the updated subset.

11. The apparatus according to claim 8 , wherein the at least two point cloud data frames include a first point cloud data frame and a second point cloud data frame, and the pose adjusting a pose parameter comprises:

determining a characteristic distance between corresponding objects in the first point cloud data frame and the second point cloud data frame; and

adjusting the pose parameter based on the characteristic distance.

12. The apparatus according to claim 8 , wherein the updating the point cloud data frame set comprises:

determining another part of the three-dimensional scene, the another part overlapping the part of the three-dimensional scene;

determining another subset corresponding to the another part of the three-dimensional scene from the point cloud data frame set; and

adjusting a pose parameter of a point cloud data frame in the subset and a pose parameter of a point cloud data frame in the another subset, to cause a third point cloud data frame in the subset to overlap a fourth point cloud data frame in the another subset.

13. The apparatus according to claim 8 , where the updating the point cloud data frame set comprises:

determining a pose parameter of a fifth point cloud data frame in the part through a global navigation satellite system; and

updating the point cloud data frame set based on the pose parameter of the fifth point cloud data frame.

14. A non-transitory computer readable storage medium, storing a computer program, wherein the program, when executed by a processor, causes the processor to perform operations, the operations comprising:

acquiring a point cloud data frame set for the three-dimensional scene, point cloud data frames in the point cloud data frame set respectively having a pose parameter;

determining a subset corresponding to a part of the three-dimensional scene from the point cloud data frame set, comprising:

determining, based on trajectories corresponding to the point cloud data frame set, a position of an intersection in the three-dimensional scene, comprising: performing dilation on a trajectory image comprising the trajectories to determine a line representing a road by dilating trajectories corresponding to the road; determining, based on the line, at least one corner point at the intersection; and determining the position of the intersection based on the at least one corner point at the intersection;

forming, based on the position of the intersection, a connection diagram connecting the intersection; and

determining, based on the connection diagram, a closed path as the part of the three-dimensional scene;

updating the subset by removing data points associated with a dynamic object from point cloud data frames in the subset;

adjusting a pose parameter of a point cloud data frame in the updated subset to obtain an adjusted subset, the adjusted subset including at least two point cloud data frames having matching overlapping parts; and

updating the point cloud data frame set using the adjusted subset.