System and method for extraction of features from a 3-D point cloud

• US 8,199,977 B2
• Filed: 05/07/2010
• Issued: 06/12/2012
• Est. Priority Date: 05/07/2010
• Status: Active Grant

• Current Assignee: Honeywell International Inc.
• Original Assignee: Honeywell International Inc.
• Inventors: Krishnaswamy, Kailash, Lukas, Jan, Kotaba, Ondrej
• Primary Examiner(s): Le, Brian Q
• Assistant Examiner(s): Bekele, Mekonen
• Application Number:
  US12/775,865
• Publication Number:
  US 20110274343A1
• Time in Patent Office: 767 Days
• Field of Search: 382/103, 382/153, 382/307, 382/204, 382/107, 382/199, 382/131, 382/106, 348/50, 348/E13.04, 345/589, 702/81
• US Class Current: 382/106
• CPC Class Codes: G06T 17/00 Three dimensional [3D] modelling, e.g. data description of 3D objects
• CPC Class Codes: G06T 2200/04 involving 3D image data
• CPC Class Codes: G06T 2207/10028 Range image; Depth image; 3D point clouds
• CPC Class Codes: G06T 7/50 Depth or shape recovery
• CPC Class Codes: G06V 10/40 Extraction of image or video features
• CPC Class Codes: G06V 20/64 Three-dimensional objects

Abstract

A method of extracting a feature from a point cloud comprises receiving a three-dimensional (3-D) point cloud representing objects in a scene, the 3-D point cloud containing a plurality of data points; generating a plurality of hypothetical features based on data points in the 3-D point cloud, wherein the data points corresponding to each hypothetical feature are inlier data points for the respective hypothetical feature; and selecting the hypothetical feature having the most inlier data points as representative of an object in the scene.

Claims

1. A method of extracting a feature from a point cloud, the method comprising:

receiving a three-dimensional (3-D) point cloud representing objects in a scene, the 3-D point cloud containing a plurality of data points;

generating a plurality of hypothetical features based on data points in the 3-D point cloud, wherein the data points corresponding to each hypothetical feature are inlier data points for the respective hypothetical feature and the respective hypothetical feature has the appearance of a geometric two-dimensional manifold, wherein the plurality of hypothetical features are planar features and the geometrical parameters for each respective planar feature comprise a normal vector and an orthogonal distance from the respective planar feature to an origin; and

selecting a hypothetical feature, which provides a value of a quality function that is closer to a desired value than the value of the quality function provided by the other hypothetical features, as representative of an object in the scene;

calculating new inlier data points based on the geometrical parameters of a fraction of the inlier data points;

wherein generating each of the plurality of hypothetical features comprises;

selecting at least three data points;

calculating geometrical parameters for the respective hypothetical feature based on the at least three data points; and

using the calculated geometrical parameters to determine the point-to-feature distance for each data point, wherein each data point whose respective point-to-feature distance is less than or equal to a first threshold value is identified as an inlier data point.

2. The method of claim 1, wherein selecting the hypothetical feature that provides the value of the quality function that is closer to the desired value comprises selecting the hypothetical feature that has the greatest number of inlier data points.

3. The method of claim 1, further comprising:

recalculating the geometrical parameters of the selected hypothetical feature based on all the inlier data points; and

identifying additional data points as inlier data points of the selected hypothetical feature based on the recalculated geometrical parameters.

4. The method of claim 3, further comprising:

determining if geometrical parameters of the selected hypothetical feature should be further recalculated based on the additional inlier data points; and

if the geometrical parameters of the selected hypothetical feature are to be further recalculated;

recalculating the geometrical parameters based on the additional inlier data points; and

using the recalculated geometrical parameters to determine the point-to-feature distance for each data point not currently identified as an inlier data point, wherein each data point whose respective point-to-feature distance is less than or equal to the first threshold value is identified as an additional inlier data point.

5. The method of claim 1, wherein calculating new inlier data points comprises:

calculating a plurality of updated hypothetical features based on the selected hypothetical feature;

selecting the updated hypothetical feature that provides the value of the quality function that is closer to the desired value than the value of the quality function provided by the other updated hypothetical features; and

recalculating the geometrical parameters of the selected updated hypothetical feature based on the additional inlier data points of the selected updated hypothetical feature;

wherein calculating each of the plurality of updated hypothetical features comprises;

randomly selecting a fraction of the inlier data points of the selected hypothetical feature;

recalculating the geometrical parameters of the selected hypothetical feature based on the fraction of the inlier data points to describe the respective updated hypothetical feature; and

using the recalculated geometrical parameters to determine the point-to feature distance for each data point not currently identified as an inlier data point, wherein each point whose respective point-to-feature distance is less than or equal to the first threshold value is identified as an additional inlier data point.

6. The method of claim 1, further comprising:

dividing the received 3-D point cloud into a plurality of cells;

wherein a plurality of hypothetical features are generated for each cell based on the data points in each respective cell and a hypothetical feature is selected from each cell.

7. The method of claim 6, further comprising:

comparing the selected hypothetical feature from each cell with the selected hypothetical feature from neighboring cells; and

if the results of the comparison are less than or equal to a merge threshold, merging the selected hypothetical features.

8. The method of claim 1, wherein selecting at least three data points further comprises:

selecting a first random data point; and

selecting data points in a neighborhood of the first random data point;

wherein the geometrical parameters are calculated if the number of selected data points is equal to or greater than a second threshold.

9. The method of claim 1, wherein geometrical parameters are calculated using a least squares fit.

10. A system comprising:

a sensor operable to obtain a set of three-dimensional (3-D) data points for a scene;

a processing unit coupled to the sensor and operable to;

generate a plurality of hypothetical features based on the plurality of data points, each hypothetical feature having a plurality of inlier data points and the appearance of a geometric two-dimensional manifold, wherein the plurality of hypothetical features are planar features and the geometrical parameters for each respective planar feature comprise a normal vector and an orthogonal distance from the respective planar feature to an origin;

select the hypothetical feature that provides a value of a quality function that is closer to a desired value than the value of the quality function provided by the other hypothetical features; and

calculate new inlier data points based on the geometrical parameters of a fraction of the plurality of inlier data points;

wherein the processing unit is further operable to generate each of the plurality of hypothetical features by;

selecting at least three data points;

calculating geometrical parameters for the respective hypothetical feature based on the at least three data points; and

calculating a point-to-feature distance for each data point based on the calculated geometrical parameters, wherein each data point whose respective point-to-feature distance is less than or equal to a first threshold value is identified as an inlier data point.

11. The system of claim 10, wherein the processing unit is further operable to:

recalculate the geometrical parameters of the selected hypothetical feature based on all the inlier data points;

identify additional data points as inlier data points of the selected hypothetical feature based on the recalculated geometrical parameters;

determine if the geometrical parameters of the selected hypothetical feature should be further recalculated based on the additional inlier data points; and

if the geometrical parameters of the selected hypothetical feature are to be further recalculated;

recalculate the geometrical parameters based on the additional inlier data points; and

determine the point-to-feature distance for each data point not currently identified as an inlier data point using the recalculated geometrical parameters, wherein each point whose respective point-to-feature distance is less than or equal to the first threshold value is identified as an inlier data point.

12. The system of claim 10, wherein the processing unit is further operable to:

calculate a plurality of updated hypothetical features based on the selected hypothetical feature;

select the updated hypothetical feature that provides a value of a quality function that is closer to the desired value than the value of the quality function provided by the other updated hypothetical features; and

recalculate the geometrical parameters of the selected updated hypothetical feature based on the additional inlier data points of the selected updated hypothetical feature;

wherein the processing unit is operable to calculate each of the plurality of updated hypothetical features by;

randomly selecting a fraction of the inliers of the selected hypothetical feature;

recalculating the geometrical parameters of the selected hypothetical feature based on the fraction of the inliers to describe the respective updated hypothetical feature; and

determining the point-to-feature distance for each data point not currently identified as an inlier data point using the recalculated geometrical parameters, wherein each point whose respective point-to-feature distance is less than or equal to the first threshold value is identified as an additional inlier data point.

13. The system of claim 10, wherein the processing unit is further operable to:

divide the plurality of data points received from the sensor into a plurality of cells, each cell comprising a subset of the plurality of data points;

generate a plurality of hypothetical features for each cell based on the subset of data points in each respective cell; and

for each cell, select the hypothetical feature that provides a value of a quality function that is closer to the desired value than the value of the quality function provided by the other hypothetical features.

14. A program product comprising program instructions embodied on a non-transitory processor-readable medium for execution by a programmable processor, wherein the program instructions are operable to cause the programmable processor to:

generate a plurality of hypothetical features based on a plurality of data points in a three-dimensional (3-D) point cloud received from a sensor, each hypothetical feature having a plurality of inlier data points and the appearance of a geometric two-dimensional manifold, wherein the plurality of hypothetical features are planar features and the geometrical parameters for each respective planar feature comprise a normal vector and an orthogonal distance from the respective planar feature to an origin;

select the hypothetical feature that provides a value of a quality function that is closer to a desired value than the value of the quality function provided by the other hypothetical planes as representative of an object feature in a scene; and

calculate new inlier data points based on the geometrical parameters of a fraction of the inlier data points;

wherein the program instructions are operable to cause the programmable processor to generate each of the plurality of hypothetical features by causing the programmable processor to;

select at least three data points;

calculate geometrical parameters for the respective hypothetical feature based on the at least three data points; and

calculate a point-to-feature distance for each data point based on the calculated geometrical parameters, wherein each data point whose respective point-to-feature distance is less than or equal to a first threshold value is identified as an inlier data point.

15. The program product of claim 14, wherein the program instructions are further operable to cause the programmable processor to:

recalculate the geometrical parameters of the selected hypothetical feature based on all the inlier data points;

identify additional data points as inlier data points of the selected hypothetical feature based on the recalculated geometrical parameters;

determine if the geometrical parameters of the selected hypothetical feature should be further recalculated based on the additional inlier data points; and

if the geometrical parameters of the selected hypothetical feature are to be further recalculated, the program instructions are further operable to cause the programmable processor to;

recalculate the geometrcial parameters based on the additional inlier datapoints; and

determine the point-to-feature distance for each data point not currently identified as an inlier data point using the recalculated geometrical parameters, wherein each data point whose respective point-to-feature distance is less than or equal to the first threshold value is identified as an additional inlier data point.

16. The program product of claim 14, wherein the program instructions are further operable to cause the programmable processor to:

calculate a plurality of updated hypothetical features based on the selected hypothetical feature;

select the updated hypothetical feature that provides the value of the quality function that is closer to the desired value than the value of the quality function provided by the other updated hypothetical features; and

recalculate the geometrical parameters of the selected updated hypothetical feature based on the additional inlier data points of the selected updated hypothetical feature;

wherein the program instructions are operable to cause the programmable processor to calculate each of the plurality of updated hypothetical features by causing the programmable processor to;

randomly select a fraction of the inliers of the selected hypothetical feature;

recalculate the geometrical parameters of the selected hypothetical feature based on the fraction of the inliers to describe the respective updated hypothetical feature; and

determine the point-to-feature distance for each data point not currently identified as an inlier data point using the recalculated geometrical parameters, wherein each data point whose respective point-to-feature distance is less than or equal to the first threshold value is identified as an additional inlier data point.

17. The program product of claim 14, wherein the program instructions are further operable to cause the programmable processor to:

divide the plurality of data points received from the sensor into a plurality of cells, each cell comprising a subset of the plurality of data points;

generate a plurality of hypothetical features for each cell based on the subset of data points in each respective cell; and

for each cell, select the hypothetical feature that provides a value of a quality function that is closer to a desired value than the value of the quality function provided by the other hypothetical features of the respective cell.