Granted Patent B2

US 12,373,913 · App. 17/940,733 · Granted Jul 29, 2025

Inventors: Zhisong Liu (Shenzhen, CN); Zijia Wang (WeiFang, CN); Zhen Jia (Shanghai, CN)

Assignee: Dell Products L.P.

G06T3/40G01S17/89

Abstract

Claims (78)

1. A method comprising:

performing upsampling on a first feature of a first point cloud of a target object;

determining a reference feature of a second point cloud of a reference object;

determining a second feature based on the first feature subjected to upsampling and the reference feature; and

generating a third point cloud of the target object based on the second feature and the second point cloud of the reference object, wherein the third point cloud has a larger number of points than the first point cloud;

wherein determining the second feature comprises:

performing multiple instances of feature extraction on the reference feature to obtain a dense feature;

performing location coding on the second point cloud to generate location codes;

determining a first parameter and a second parameter based on the dense feature and the location codes; and

determining the second feature based on the first parameter, the second parameter, and the first feature subjected to upsampling.

2. The method according to claim 1 , further comprising:

generating a fourth point cloud based on the first feature subjected to upsampling and the third point cloud, wherein the fourth point cloud has the same number of points as the third point cloud.

3. The method according to claim 1 , wherein performing upsampling on the first feature of the first point cloud of the target object comprises:

performing multiple instances of sampling on the first feature so as to make a dimension of the first feature be the same as a dimension of the reference feature.

4. The method according to claim 1 , further comprising:

scanning the target object by a hardware device to obtain the first point cloud; and

determining the first feature based on the first point cloud by using a machine learning model.

5. The method according to claim 4 , wherein the hardware device comprises a laser radar.

6. The method according to claim 1 , wherein determining the second feature further comprises:

determining the reference feature based on the second point cloud by using a machine learning model.

7. The method according to claim 6 , wherein performing multiple instances of feature extraction on the reference feature to obtain the dense feature comprises:

determining a nearest neighbor feature based on the reference feature, wherein a distance between the nearest neighbor feature and the reference feature is within a threshold value;

determining a first intermediate feature based on the nearest neighbor feature by using the machine learning model;

combining the first intermediate feature and the nearest neighbor feature to determine a second intermediate feature; and

determining the dense feature based on the second intermediate feature by using the machine learning model.

8. The method according to claim 1 , wherein determining the first parameter and the second parameter based on the dense feature and the location codes comprises:

combining the dense feature and the location codes to determine a combined feature; and

determining the first parameter and the second parameter based on the combined feature by using an attention mechanism of a machine learning model.

9. An electronic device, comprising:

at least one processor; and

memory coupled to the at least one processor, wherein the memory has instructions stored therein which, when executed by the at least one processor, cause the electronic device to perform actions comprising:

performing upsampling on a first feature of a first point cloud of a target object;

determining a reference feature of a second point cloud of a reference object;

determining a second feature based on the first feature subjected to upsampling and the reference feature; and

generating a third point cloud of the target object based on the second feature and the second point cloud of the reference object, wherein the third point cloud has a larger number of points than the first point cloud;

wherein determining the second feature comprises:

performing multiple instances of feature extraction on the reference feature to obtain a dense feature;

performing location coding on the second point cloud to generate location codes;

determining a first parameter and a second parameter based on the dense feature and the location codes; and

determining the second feature based on the first parameter, the second parameter, and the first feature subjected to upsampling.

10. The electronic device according to claim 9 , wherein the actions further comprise:

generating a fourth point cloud based on the first feature subjected to upsampling and the third point cloud, wherein the fourth point cloud has the same number of points as the third point cloud.

11. The electronic device according to claim 9 , wherein performing upsampling on the first feature of the first point cloud of the target object comprises:

performing multiple instances of sampling on the first feature so as to make a dimension of the first feature be the same as a dimension of the reference feature.

12. The electronic device according to claim 9 , wherein the actions further comprise:

scanning the target object by a hardware device to obtain the first point cloud; and

determining the first feature based on the first point cloud by using a machine learning model.

13. The electronic device according to claim 12 , wherein the hardware device comprises a laser radar.

14. The electronic device according to claim 9 , wherein determining the second feature further comprises:

determining the reference feature based on the second point cloud by using a machine learning model.

15. The electronic device according to claim 14 , wherein performing multiple instances of feature extraction on the reference feature to obtain the dense feature comprises:

determining a nearest neighbor feature based on the reference feature, wherein a distance between the nearest neighbor feature and the reference feature is within a threshold value;

determining a first intermediate feature based on the nearest neighbor feature by using the machine learning model;

combining the first intermediate feature and the nearest neighbor feature to determine a second intermediate feature; and

determining the dense feature based on the second intermediate feature by using the machine learning model.

16. The electronic device according to claim 9 , wherein determining the first parameter and the second parameter based on the dense feature and the location codes comprises:

combining the dense feature and the location codes to determine a combined feature; and

determining the first parameter and the second parameter based on the combined feature by using an attention mechanism of a machine learning model.

17. A computer program product tangibly stored on a non-transitory computer-readable medium and including computer-executable instructions that, when executed by a device, cause the device to perform actions comprising:

performing upsampling on a first feature of a first point cloud of a target object;

determining a reference feature of a second point cloud of a reference object;

determining a second feature based on the first feature subjected to upsampling and the reference feature; and

generating a third point cloud of the target object based on the second feature and the second point cloud of the reference object, wherein the third point cloud has a larger number of points than the first point cloud;

wherein determining the second feature comprises:

performing multiple instances of feature extraction on the reference feature to obtain a dense feature;

performing location coding on the second point cloud to generate location codes;

determining a first parameter and a second parameter based on the dense feature and the location codes; and

determining the second feature based on the first parameter, the second parameter, and the first feature subjected to upsampling.

18. The computer program product according to claim 17 , wherein the actions further comprise:

generating a fourth point cloud based on the first feature subjected to upsampling and the third point cloud, wherein the fourth point cloud has the same number of points as the third point cloud.

19. The computer program product according to claim 17 , wherein performing multiple instances of feature extraction on the reference feature to obtain the dense feature comprises:

determining a nearest neighbor feature based on the reference feature, wherein a distance between the nearest neighbor feature and the reference feature is within a threshold value;

determining a first intermediate feature based on the nearest neighbor feature by using a machine learning model;

combining the first intermediate feature and the nearest neighbor feature to determine a second intermediate feature; and

determining the dense feature based on the second intermediate feature by using the machine learning model.

20. The computer program product according to claim 17 , wherein determining the first parameter and the second parameter based on the dense feature and the location codes comprises:

combining the dense feature and the location codes to determine a combined feature; and

determining the first parameter and the second parameter based on the combined feature by using an attention mechanism of a machine learning model.

Assignments (1)

Priority Claims (1)

Continuity (1)

References Cited (28)