Granted Patent B2

US 12,387,285 · App. 18/004,307 · Granted Aug 12, 2025

Inventors: Lei Yang (Beijing, CN); Kai Wu (Beijing, CN)

Assignee: BEIJING JINGDONG QIANSHI TECHNOLOGY CO., LTD.

G06T1/0007G06T1/20G06T1/60G06T7/70G06V10/46G06T2207/10028G06T2207/30196

Abstract

Claims (61)

1. An object detection method, comprising:

obtaining raw point cloud data comprising a to-be-detected object, wherein the to-be-detected object is annotated in the raw point cloud data so that the raw point cloud data comprises annotation information of the to-be-detected object;

extracting, from the raw point cloud data, instance point cloud data corresponding to the to-be-detected object by using the annotation information, wherein the instance point cloud data is data of a point cloud corresponding to the to-be-detected object in the raw point cloud data;

determining an object position point from the raw point cloud data, and fusing the instance point cloud data into the object position point of the raw point cloud data to obtain a fused to-be-detected sample, wherein the object position point is a point in a ground point cloud of the raw point cloud data; and

detecting the to-be-detected object by using the raw point cloud data and the to-be-detected sample.

2. The method according to claim 1 , wherein the fusing the instance point cloud data into the object position point of the raw point cloud data comprises:

performing translation transformation on the instance point cloud data according to the object position point to obtain fused point cloud data as the to-be-detected sample.

3. The method according to claim 1 , wherein the determining the object position point from the raw point cloud data comprises:

performing a ground detection on the raw point cloud data to determine a ground point cloud in the raw point cloud data; and

outputting a candidate position point from the ground point cloud, and determining the candidate position point as the object position point when the candidate position point satisfies a collision detection condition.

4. The method according to claim 3 , wherein when the instance point cloud data is positioned at the candidate position point and does not collide with an object other than the ground point cloud in the raw point cloud data, the candidate position point satisfies the collision detection condition.

5. The method according to claim 2 , further comprising, after obtaining the fused point cloud data:

calculating an occluded point cloud data between to-be-fused point cloud data and the raw point cloud data; and

removing the occluded point cloud data from the fused point cloud data to obtain the to-be-detected sample.

6. The method according to claim 5 , wherein the calculating the occluded point cloud data between the fused point cloud data and the raw point cloud data comprises:

projecting the fused point cloud data into a spherical coordinate system to obtain a corresponding spherical projection; and

sampling the fused point cloud data, and when a corresponding pixel point of the sampled sample point in the spherical projection is not empty, determining the occluded point cloud data according to the sample point.

7. The method according to claim 6 , wherein the determining the occluded point cloud data according to the sample point comprises:

obtaining a first distance of the sample point from an origin of the spherical coordinate system and a second distance of the corresponding pixel point of the sample point in the spherical projection from the origin of the spherical coordinate system; and

determining the sample point as the occluded point cloud data when the first distance is greater than the second distance.

8. An object detection device, comprising:

a processor; and

a memory for storing executable instructions that, when being executed by the processor, cause the processor to implement an object detection method comprising:

obtaining raw point cloud data comprising a to-be-detected object, wherein the to-be-detected object is annotated in the raw point cloud data so that the raw point cloud data comprises annotation information of the to-be-detected object;

extracting, from the raw point cloud data, instance point cloud data corresponding to the to-be-detected object by using the annotation information, wherein the instance point cloud data is data of a point cloud corresponding to the to-be-detected object in the raw point cloud data;

determining an object position point from the raw point cloud data, and fusing the instance point cloud data into the object position point of the raw point cloud data to obtain a fused to-be-detected sample, wherein the object position point is a point in a ground point cloud of the raw point cloud data; and

detecting the to-be-detected object by using the raw point cloud data and the to-be-detected sample.

9. The device according to claim 8 , wherein the fusing the instance point cloud data into the object position point of the raw point cloud data comprises:

performing translation transformation on the instance point cloud data according to the object position point to obtain fused point cloud data as the to-be-detected sample.

10. The device according to claim 8 , wherein the determining the object position point from the raw point cloud data comprises:

performing a ground detection on the raw point cloud data to determine a ground point cloud in the raw point cloud data; and

outputting a candidate position point from the ground point cloud, and determining the candidate position point as the object position point when the candidate position point satisfies a collision detection condition.

11. The device according to claim 10 , wherein

when the instance point cloud data is positioned at the candidate position point and does not collide with an object other than the ground point cloud in the raw point cloud data, the candidate position point satisfies the collision detection condition.

12. The device according to claim 9 , wherein the object detection method further comprises, after obtaining the fused point cloud data:

calculating an occluded point cloud data between to-be-fused point cloud data and the raw point cloud data; and

removing the occluded point cloud data from the fused point cloud data to obtain the to-be-detected sample.

13. The device according to claim 12 , wherein the calculating the occluded point cloud data between the fused point cloud data and the raw point cloud data comprises:

projecting the fused point cloud data into a spherical coordinate system to obtain a corresponding spherical projection; and

sampling the fused point cloud data, and when a corresponding pixel point of the sampled sample point in the spherical projection is not empty, determining the occluded point cloud data according to the sample point.

14. The device according to claim 13 , wherein the determining the occluded point cloud data according to the sample point comprises:

obtaining a first distance of the sample point from an origin of the spherical coordinate system and a second distance of the corresponding pixel point of the sample point in the spherical projection from the origin of the spherical coordinate system; and

determining the sample point as the occluded point cloud data when the first distance is greater than the second distance.

15. A computer-readable non-volatile storage media having executable instructions stored thereon that, when being executed by a processor, cause the processor to implement an object detection method comprising:

obtaining raw point cloud data comprising a to-be-detected object, wherein the to-be-detected object is annotated in the raw point cloud data so that the raw point cloud data comprises annotation information of the to-be-detected object;

extracting, from the raw point cloud data, instance point cloud data corresponding to the to-be-detected object by using the annotation information, wherein the instance point cloud data is data of a point cloud corresponding to the to-be-detected object in the raw point cloud data;

determining an object position point from the raw point cloud data, and fusing the instance point cloud data into the object position point of the raw point cloud data to obtain a fused to-be-detected sample, wherein the object position point is a point in a ground point cloud of the raw point cloud data; and

detecting the to-be-detected object by using the raw point cloud data and the to-be-detected sample.

16. The storage media according to claim 15 , wherein the fusing the instance point cloud data into the object position point of the raw point cloud data comprises:

performing translation transformation on the instance point cloud data according to the object position point to obtain fused point cloud data as the to-be-detected sample.

17. The storage media according to claim 15 , wherein the determining the object position point from the raw point cloud data comprises:

performing a ground detection on the raw point cloud data to determine a ground point cloud in the raw point cloud data; and

outputting a candidate position point from the ground point cloud, and determining the candidate position point as the object position point when the candidate position point satisfies a collision detection condition.

18. The storage media according to claim 17 , wherein

when the instance point cloud data is positioned at the candidate position point and does not collide with an object other than the ground point cloud in the raw point cloud data, the candidate position point satisfies the collision detection condition.

19. The storage media according to claim 16 , wherein the object detection method further comprises, after obtaining the fused point cloud data:

calculating an occluded point cloud data between to-be-fused point cloud data and the raw point cloud data; and

removing the occluded point cloud data from the fused point cloud data to obtain the to-be-detected sample.

20. The storage media according to claim 19 , wherein the calculating the occluded point cloud data between the fused point cloud data and the raw point cloud data comprises:

projecting the fused point cloud data into a spherical coordinate system to obtain a corresponding spherical projection; and

sampling the fused point cloud data, and when a corresponding pixel point of the sampled sample point in the spherical projection is not empty, determining the occluded point cloud data according to the sample point.

Assignments (1)

Priority Claims (1)

Continuity (1)

References Cited (17)