System for determining the pose of an object which utilizes range profiles and synethic profiles derived from a model

• US 5,471,541 A
• Filed: 11/16/1993
• Issued: 11/28/1995
• Est. Priority Date: 11/16/1993
• Status: Expired due to Fees

• Current Assignee: National Research Council Canada
• Original Assignee: National Research Council Canada
• Inventors: Burtnyk, Nestor, Greenspan, Michael A.
• Primary Examiner(s): NOT, DEFINED
• Assistant Examiner(s): Chang, Jon
• Application Number:
  US08/152,745
• Time in Patent Office: 742 Days
• Field of Search: 382/8, 382/1, 382/48, 382/30, 364/559, 348/94, 395/94, 356/375, 356/376
• US Class Current: 382/153
• CPC Class Codes: G06T 7/75 involving models

Abstract

A method for determining the 3-D pose of an object is provided assuming that an approximate pose of the object is known. As a laser range camera scans the object, depth profiles of the object are obtained and stored in a memory. Synthetic profiles are obtained from a model of the object that has been oriented to the assumed pose of the object, by using a conventional raster rendering technique to obtain a series of slices of the model. Once this synthetic profile signature of the model of the object is generated, a comparison is performed between this profile signature and the profiles obtained from the laser range camera. When a suitable match is found, a correction to the 3-D pose is calculated from the offset parameters associated with the match; further refinement may be performed by repeating the process.

Claims

1. A method of determining the 3-D position and orientation of an object relative to a scanning device, given a surface model of the object, range profiles that have been acquired by scanning the object, and the object'"'"'s approximate location defined by the 3-D position and orientation relative to the scanning device, the method comprising the steps of:

a) generating a set of at least 2 synthetic 2-D profiles of the object from a model of the object by simulating how the scanning device would have scanned the object if the object were located at the approximate 3-D location;

b) searching through the acquired range profiles to select a subset of these profiles whose 2-D waveforms most closely match the 2-D waveforms of the set of synthetic profiles;

c) determining the amount of 2-D translational and rotational offset between the 2-D waveforms of the synthetic profiles and the 2-D waveforms of the acquired range profiles that produced this most close match;

d) calculating a correction of the approximate three dimensional position and orientation of the model representing the object from these 2-D offsets between the synthetic profiles and the acquired range profiles that produced the best match in step (b) to obtain a closer approximate 3-D position and orientation of the object; and

,

e) repeating steps (a) to (d) until the correction is below a predetermined threshold.

2. A method as defined in claim 1, wherein the generated synthetic profiles are comprised of a plurality of data points, the profile being a cross-section of the model of the object generated as a slice through a fractional point along an axis of the model.

3. A method as defined in claim 1, wherein the step of searching includes filtering comprising the steps of:

segmenting the acquired range profiles into line segments characterizing their coarse shapes;

storing the line segments in a random access memory;

segmenting the at least two synthetic profiles into line segments characterizing their coarse shapes;

comparing the coarse segments of the at least two synthetic profiles with the coarse segments of the acquired range profiles to obtain candidate matches; and

evaluating each candidate further by comparing the waveform shapes of the at least two synthetic profiles with the data points of the corresponding acquired range profiles to obtain a best match.

4. A method as defined in claim 3, further comprising the steps of:

(i) selecting a synthetic profile that has the greatest deviation from a straight line as a principal synthetic profile;

(ii) locating a segment in the principal synthetic profile that has the longest horizontal span, that segment being a seed segment; and

,

(iii) for each segmented range profile, searching for an acceptable candidate match by aligning the seed segment with each data segment, wherein steps (i) to (iii) are performed after the step of segmenting in claim 3.