Voting-based pose estimation for 3D sensors

• US 8,908,913 B2
• Filed: 12/19/2011
• Issued: 12/09/2014
• Est. Priority Date: 12/19/2011
• Status: Active Grant

• Current Assignee: Mitsubishi Electric Research Laboratories, Inc. (Mitsubishi Electric Corporation)
• Original Assignee: Mitsubishi Electric Research Laboratories, Inc. (Mitsubishi Electric Corporation)
• Inventors: Taguchi, Yuichi, Tuzel, Oncel, Ramalingam, Srikumar, Choi, Changhyun, Liu, Ming-Yu
• Primary Examiner(s): Bitar, Nancy
• Application Number:
  US13/329,493
• Publication Number:
  US 20130156262A1
• Time in Patent Office: 1,086 Days
• Field of Search: None
• US Class Current: 382/103
• CPC Class Codes: G06T 2207/30108 Industrial image inspection
• CPC Class Codes: G06T 2207/30164 Workpiece; Machine component
• CPC Class Codes: G06T 7/521 from laser ranging, e.g. using interferometry; from the projection of structured light
• CPC Class Codes: G06T 7/75 involving models

Abstract

A pose of an object is estimated by first defining a set of pair features as pairs of geometric primitives, wherein the geometric primitives include oriented surface points, oriented boundary points, and boundary line segments. Model pair features are determined based on the set of pair features for a model of the object. Scene pair features are determined based on the set of pair features from data acquired by a 3D sensor, and then the model pair features are matched with the scene pair features to estimate the pose of the object.

Claims

1. A method for estimating a pose of an object, comprising the steps of:

defining a set of pair features as pairs of geometric primitives defined in a 3D space invariant to the poses of the object, wherein the geometric primitives include oriented surface points, oriented boundary points, and boundary line segments;

determining model pair features based on the set of pair features for a model of the object;

determining scene pair features based on the set of pair features from data acquired by a 3D sensor;

determining a descriptor from each pair feature, and matching the descriptors model pair features with the descriptors scene pair features;

discretizing the descriptors for the pair features;

storing the discretized descriptors of the model pair features; and

performing the matching using a comparison algorithm, estimating candidate poses of the object based on voting via an intermediate coordinate transformation, wherein the steps are performed in a processor.

2. The method of claims 1, wherein the object is one of a plurality of different object randomly commingled with other objects, and there are model pair features for each object.

3. The method of claims 1, wherein the 3D sensor is mounted on a robotic arm, and the robotic arm includes a gripper to pick the object according to the pose.

4. The method of claims 1, wherein the 3D sensor uses structured light generated by an infrared laser.

5. The method of claim 1, wherein each pair feature is defined using two oriented surface points, two oriented boundary points, one oriented surface point and one oriented boundary point, and two boundary line segments.

6. The method of claim 1, wherein each oriented surface point comprises a 3D position on a surface of the object and a 3D normal vector.

7. The method of claim 1, wherein each oriented boundary point comprises a 3D position on a boundary of the object and a 3D direction vector.

8. The method of claim 7, wherein the 3D direction vector is defined as a direction tangent to the boundary at the 3D position.

9. The method of claim 1, wherein each boundary line segment is a line segment fitted to 3D points on a boundary of the object.

10. The method of claims 1, further comprising:

storing the discretized descriptors of the model pair features in a hash table; and

performing the matching using the hash table.

11. The method of claim 1, further comprising:

clustering the candidate poses.

12. The method of claim 1, further comprising:

refining the pose using an iterative-closest point procedure.