METHOD OF SEPARATING OBJECT IN THREE DIMENSION POINT CLOUD

• US 20130051658A1
• Filed: 08/14/2012
• Published: 02/28/2013
• Est. Priority Date: 08/22/2011
• Status: Active Grant

• Current Assignee: Samsung Electronics Co. Ltd.
• Original Assignee: Samsung Electronics Co. Ltd.
• Inventors: Hwang, Hyo Seok, Roh, Kyung Shik, Yoon, Suk June
• Granted Patent: US 9,251,399 B2
• Time in Patent Office: Days
• US Class Current: 382/154
• CPC Class Codes: G06T 2207/10028 Range image; Depth image; 3D point clouds
• CPC Class Codes: G06T 7/11 Region-based segmentation
• CPC Class Codes: G06T 7/187 involving region growing; involving region merging; involving connected component labelling
• CPC Class Codes: G06V 20/64 Three-dimensional objects

Abstract

A method of separating an object in a three dimension point cloud including acquiring a three dimension point cloud image on an object using an image acquirer, eliminating an outlier from the three dimension point cloud image using a controller, eliminating a plane surface area from the three dimension point cloud image, of which the outlier has been eliminated using the controller, and clustering points of an individual object from the three dimension point cloud image, of which the plane surface area has been eliminated using the controller.

Claims

1. A method of separating an object in a three dimension point cloud, the method comprising:

using an image acquirer, acquiring a three dimension point cloud image on an object;

using a controller, eliminating an outlier from the three dimension point cloud image;

using a controller, eliminating a plane surface area from the three dimension point cloud image, of which the outlier has been eliminated; and

using the controller, clustering points of an individual object from the three dimension point cloud image, of which the plane surface area has been eliminated.

2. The method of claim 1, wherein the image acquirer comprises a stereo camera, a Time of Flight (TOF) camera, a Laser Range Finders (LRF) sensor, or a kinect sensor.

3. The method of claim 1, wherein in the eliminating of the outlier, whether a distance between each point of the three dimension point cloud image and nearby points adjacent to the each point is equal to or greater than a critical value is determined.

4. The method of claim 3, wherein the eliminating of the outlier comprises:

extracting a predetermined number of nearest points with respect to each point of the three dimension point cloud image;

measuring an average distance between the each point and the nearest points; and

eliminating the each point and the nearest points if an average distance between the each point and the nearest points is equal to or greater than a critical value.

5. The method of claim 1, wherein in the eliminating of the outlier, if a distance between each point of the three dimension point cloud image and a position at which the three dimension point cloud image on the object is acquired is equal to or greater than a predetermined distance, the controller eliminates the each point.

6. The method of claim 1, wherein in the eliminating of the plane surface area, whether a distance between each point and a plane surface of the three dimension point cloud image is equal to or greater than a critical value is determined.

7. The method of claim 6, wherein the eliminating of the plane surface area comprises:

extracting a predetermined number of plane surfaces, each plane surface passing through three points on the three dimension point cloud image, and measuring the number of points having a distance that is less than a critical value with respect to each plane surface on the three dimension point cloud image;

selecting a plane surface having a largest number of points having a distance less than the critical value with respect to each plane surface; and

eliminating each point in case when a distance between the each point on the three dimension point cloud image and the selected plane surface is less than the critical value.

8. The method of claim 1, wherein the clustering of points of an individual object comprises:

assigning a label to each point on the three dimension point cloud image; and

matching each point cloud segment assigned with same label, thereby calibrating to a point cloud of an individual object.

9. The method of claim 8, wherein the assigning a label to each point comprises:

extracting adjacent points that are within a predetermined distance with respect to each point on the three dimension point cloud image; and

assigning a same label to the each point and to the adjacent points if the number of the adjacent points is equal to or greater than a critical value.

10. The method of claim 8, wherein in the calibrating to a point cloud of an individual object, if a label of each point on the three dimension point cloud image is different from a label of an adjacent point that is within a predetermined distance with respect to the each point, the controller assigns the each point on the three dimension point cloud image and the adjacent point with same label.

11. The method of claim 1, wherein in the eliminating of the outliner, a downsampling is conducted on the three dimension point cloud image at a predetermined rate, and an outlier is eliminated from the three dimension point cloud image which has been downsampled.

12. The method of claim 11, wherein in the eliminating of the outlier, the controller determines if the distances between each point on the three dimension point cloud image and points adjacent to the each point are equal to or greater than a critical value.

13. The method of claim 12, wherein the eliminating of the outlier comprises:

extracting a predetermined number of nearest points with respect to each point of the three dimension point cloud image;

measuring an average distance between the each point and the nearest points; and

eliminating the each point and the nearest points if the average distance is equal to or greater than a critical value.

14. The method of claim 11, wherein in the eliminating of the outlier, if a distance between each point on the three dimension point cloud image and a position at which the three dimension point cloud image on the object is acquired is equal to or greater than a predetermined distance, the controller eliminates the each point.

15. The method of claim 11, wherein in the eliminating of the plane surface area, the controller determines whether a distance between each point and a plane surface of the three dimension point cloud image is equal to or greater than a critical value.

16. The method of claim 15, wherein the eliminating the plane surface area comprises:

extracting a predetermined number of plane surfaces, each plane surface passing through three points on the three dimension point cloud image, and measuring the number of points having a distance that is less than a critical value with respect to each plane surface on the three dimension point cloud image;

selecting a plane surface having a largest number of points having a distance less than the critical value with respect to each plane surface; and

in case when a distance between each point on the three dimension point cloud image and the selected plane surface is less than a critical value.

17. The method of claim 11, wherein the clustering of points of an individual object comprises:

assigning a label to each point on the three dimension point cloud image; and

matching each point cloud segment assigned with same label, thereby calibrating to a point cloud of an individual object.

18. The method of claim 17, wherein the assigning of a label to each point comprises:

extracting adjacent points that are within a predetermined distance with respect to each point on the three dimension point cloud image; and

assigning a same label to the each point and to the adjacent points if the number of the adjacent points is equal to or greater than a critical value.

19. The method of claim 17, wherein in the calibrating to a point cloud of an individual object, if a label of each point on the three dimension point cloud image is different from a label of an adjacent point that is within a predetermined distance with respect to the each point, the controller assigns the each point and the adjacent point with same label.

20. At least one non-transitory computer readable medium storing computer readable instructions that control at least one processor to perform a method comprising:

acquiring a three dimension point cloud image on an object;

eliminating an outlier from the three dimension point cloud image;

eliminating a plane surface area from the three dimension point cloud image, of which the outlier has been eliminated; and

clustering points of an individual object from the three dimension point cloud image, of which the plane surface area has been eliminated.