See discussions, stats, and author profiles for this publication at: https://www.researchgate.

net/publication/260320178

NARF: 3D Range Image Features for Object Recognition

Article · January 2010

CITATIONS READS

83 915

4 authors, including:

Bastian Steder Wolfram Burgard

University of Freiburg University of Freiburg
39 PUBLICATIONS   931 CITATIONS    625 PUBLICATIONS   39,162 CITATIONS   

SEE PROFILE SEE PROFILE

Some of the authors of this publication are also working on these related projects:

Flourish View project

NeuroBots View project

All content following this page was uploaded by Bastian Steder on 14 January 2015.

The user has requested enhancement of the downloaded file.

 NARF: 3D Range Image Features for Object Recognition
Bastian Steder Radu Bogdan Rusu Kurt Konolige Wolfram Burgard

Abstract— We present our findings regarding a novel method

for interest point detection and feature descriptor calculation in
3D range data called NARF (Normal Aligned Radial Feature).
The method makes explicit use of object boundary information
and tries to extract the features in areas where the surface is
stable but has substantial change in the vicinity.
(a)
I. I NTRODUCTION
In object recognition or mapping applications, the ability
to find similar parts in different sets of sensor readings is
very important. Most methods use features that describe
a chunk of data in a compressed representation and that
(b)
can be used to efficiently perform comparisons between
different data regions. The task mainly consists of two parts,
the identification of appropriate points, often referred to as
interest points or key points, and the way in which the
information in the vicinity of that point is encoded in a
descriptor or description vector. (c)
In this work we focus on single range scans, as obtained
with 3D laser range finders or stereo cameras, where the
data is incomplete and dependent on a viewpoint. We chose
range images as the way to represent the data since they
reflect this situation and enable us to borrow ideas from the
vision sector. (d)
We present the normal aligned radial feature (NARF), a
novel interest point extraction method together with a feature Fig. 1. The interest point extraction procedure. (a) Range image of an
office scene with two chairs in the front with the extracted borders marked.
descriptor for points in 3D range data. The interest point (b) Surface change scores according to borders and principle curvature. (c)
extraction method has been designed with two specific goals Interest values with marked interest points for a support size of 20cm. Note
in mind: i) the selected points are in positions where the how the corners of the chairs are detected as interest points at this scale. (d)
Interest values with marked interest points for a support size of 1m. Note
surface is stable (to ensure a robust estimation of the normal) how, compared to (c), the whole surface of the chair’s backrests contain one
and where there are sufficient changes in the immediate interest point at this scale.
vicinity; ii) since we focus on partial views, we want to make
use of object borders, meaning the outer shapes of objects
seen from a certain perspective. The outer forms are often attention on significant structures. We have the following
rather unique so that their explicit use in the interest point requirements for our interest point extraction procedure:
extraction and the descriptor calculation can be expected to i) it must take information about borders and the surface
make the overall process more robust. structure into account; ii) it must select positions that can be
Figure 1 shows an outline of our interest point extraction reliably detected even if the object is observed from another
procedure. The implementation of our method is available perspective; and iii) the points must be on positions that
under an open-source license1 . provide stable areas for normal estimation or the descriptor
calculation in general.
II. NARF I NTEREST POINT EXTRACTION Stable interest points need significant changes of the
The detection of interest points is an important step to surface in a local neighborhood to be robustly detected in
reduce the search space for feature extraction and focus the the same place even if observed from different perspectives.
This typically means, that there are substantially different
B. Steder and W. Burgard are with the Dept. of Computer Science dominant directions of the surface changes in the area. To
of the University of Freiburg, Germany. {steder,burgard}@informatik.uni-
freiburg.de; R. Rusu and K. Konolige are with Willow Garage Inc., USA. capture this, we
{rusu,konolige}@willowgarage.com
1 This work is part of the ROS point cloud library. See • find borders in the range image, meaning non-
http://www.ros.org/wiki/pcl/Tutorials continuous traversals from foreground to background,
 by looking for substantial increases in the 3D distances
between neighboring image points.
• look at the local neighborhood of every image point
and determine a score how much the surface changes
at this position and a dominant direction for this change,
incorporating the information about borders,
(a)
• look at the dominant directions in the surrounding of
each image point and calculate an interest value that
represents i) how much these directions differ from each
other and ii) how much the surface in the point itself
changes (meaning how stable it is),
• perform smoothing on the interest values, and
• perform non-maximum suppression to find the final (c) (b)
interest points.
Fig. 2. (a): A range image of an example scene with an armchair in
the front. The black cross marks the position of an interest point. (b):
III. T HE NARF DESCRIPTOR Visualization how the descriptor is calculated. The top shows a range value
Feature descriptors describe the area around an interest patch of the top right corner of the armchair. The actual descriptor is
visualized on the bottom. Each of the 20 cells of the descriptor corresponds
point in a way that makes efficient comparison regarding to one of the beams (green) visualized in the patch, with two of the
similarity possible. Our goals in the development for the correspondences marked with arrows. The additional (red) arrow pointing to
NARF descriptor were i) that it captures the existence of the top right shows the extracted dominant orientation. (c): The descriptor
distances to every other point in the scene (the brighter the higher the value).
occupied and free space, so that parts on the surface and Note that mainly top right rectangular corners get low values.
also the outer shape of an object can be described, ii) that
it is robust against noise on the interest point position, and 1
iii) that it enables us to extract a unique local coordinate
frame at the point.
0.1
To compute the NARF descriptor in an interest point, we True positive rate

• calculate a normal aligned range value patch in the

point, which is a small range image with the observer 0.01

looking at the point along the normal,

• overlay a star pattern onto this patch, where each beam 0.001 NARFs all points
NARFs int. points
corresponds to a value in the final descriptor, that NARFs(ri) int. points
RVPs int. points
captures how much the pixels under the beam change, 0.0001
• extract a unique orientation from the descriptor, 0.0001 0.001 0.01 0.1 1
False positive rate
• and shift the descriptor according to this value to make
it invariant to the rotation. Fig. 3. This graph shows ROC curves for feature matches where the
threshold for the descriptor distance increases from the bottom left to the
The unique orientation together with the normal then defines top right. Please note, that the axes are logarithmic. The higher a line in
a local 6DOF coordinate frame at the point. Please consider this graph, the better is the performance.
Figure 2 for a visual example of the process and application.
IV. R ESULTS is lower than 1:10. The plot marked NARFs all points is
We wanted to test the capability of our NARF-descriptors for our NARF feature descriptors extracted at every image
to match features from an object model to the features in a point, without using the interest points. The plot marked
scene containing the model. Figure 3 shows the results for NARFs int. points shows the performance of the NARF
the matching process of seven object models (armchair, cart, features together with the interest points. The plot marked
chair, cup, pioneer robot, stapler, and teddy bear) against NARFs(ri) int. points is for the rotational invariant version
scenes with known ground truth positions. To match the of the NARF descriptor. The plot marked RVPs int. points
objects against the scenes, we sampled poses from our is using the range value patches that we used in our former
models that differed from the ground truth poses between work [1] as feature descriptors. As can be seen, the interest
0◦ and 50◦ . points are a definite improvement compared to random point
The results of true positives and false positives are sum- positions. Additionally, the rotationally invariant version and
marized as ROC (Relative Operating Characteristic) curves. the rotationally variant version of the NARF descriptor
Please note the logarithmic scales, which show the interesting outperform the RVPs we used before.
part of the plot, the bottom left corner, better. The absolute R EFERENCES
number of false positives is much higher than the absolute
[1] B. Steder, G. Grisetti, M. Van Loock, and W. Burgard. Robust on-
number of true positives, which makes areas with a high line model-based object detection from range images. In Proc. of
ratio of false positives less useful. The thicker plots mark the IEEE/RSJ Int. Conf. on Intelligent Robots and Systems (IROS), St.
areas, where the number of true positives to false positives Louis, MO, USA, October 2009.

View publication stats