Int J Adv Manuf Technol (****) **:*** ***

DOI **.1007/s00170-009-1951-9

ORIGINAL ARTICLE

Planar segmentation of data from a laser pro le scanner

mounted on an industrial robot

J. A. P. Kjellander Mohamed Rahayem

Received: 29 October 2008 / Accepted: 27 January 2009 / Published online: 24 February 2009

Springer-Verlag London Limited 2009

Abstract In industrial applications like rapid prototyp- cover the entire object the pro le scanner is moved

ing, robot vision, and geometric reverse engineering, along the surface and new images are captured. The

where speed and automatic operation are important, result is a series of pro les, each captured with the

an industrial robot and a laser pro le scanner can scanner in a different pose. The pro le scanner is usu-

be used as a 3D measurement system. This paper is ally mounted on a mechanical device that controls the

concerned with the problem of segmenting the data scanner movement, or at least records the scanner pose

from such a system into regions that can be tted with of each pro le in 3D. This makes it possible to map the

planar surfaces. We have developed a new algorithm 2D points in the pro les to a common 3D coordinate

for planar segmentation based on laser scan pro les and system and the result is then often referred to as a

robot poses. Compared to a traditional algorithm that point cloud. Point clouds may be used in applications

operates on a point cloud, the new algorithm is shown like inspection, geometric reverse engineering, object

to be more effective and faster. recognition or navigation. In such applications the point

cloud is often processed by a segmentation algorithm

Keywords 3D measurement system that uses a geometrical constraint to group points into

Laser pro le scanner Industrial robot regions representing planes, cylinders or higher order

Segmentation Planar regions surfaces. Planar segmentation is thus de ned as the

problem of identifying points that belong to the same

plane. We believe that the industrial robot, although

1 Introduction not widely used for this purpose, is an interesting al-

ternative as a carrier of a laser pro le scanner. A robot

Optical measurement systems can be used to rapidly is fast, exible, robust and relatively cheap compared

acquire the coordinates of dense sets of 3D points from with a coordinate measuring machine (CMM). To test

the surfaces of real world objects. One such system is the idea, we have mounted a laser pro le scanner on

the laser pro le scanner, see Fig. 1, which projects a an industrial robot with a turntable, see Fig. 2 and see

straight line on the object while a digital camera cap- [17 19] for details on motion control, image processing,

tures the image of the projection. Pixels representing and noise ltering. It is important to note that the

the projected line are then joined into a 2D pro le. To absolute accuracy of an industrial robot is much lower

than the accuracy of a laser pro le scanner. In [27],

we show that the accuracy of our robot is in the range

J. A. P. Kjellander M. Rahayem (B) of 400 m, while the accuracy of the laser pro le

School of Science and Technology, rebro University, scanner is approx. 10 times better ( 50 m). Individual

SE-701 82 rebro, Sweden

pro les will thus be relatively accurate, but accuracy is

e-mail: abpxte@r.postjobfree.com

lost when they are mapped to the point cloud. In the

URL: www.oru.se/nt/cad

scope of planar segmentation, it should therefore be of

J. A. P. Kjellander

interest to investigate if segmentation algorithms can

e-mail: abpxte@r.postjobfree.com

182 Int J Adv Manuf Technol (2009) 45:181 190

Fig. 1 The laser pro le scanner mounted on an industrial robot

take advantage of the relatively high accuracy of the 2D

pro les. It is also likely that an algorithm operating on

2D data would be faster than an algorithm based on 3D

point clouds. This has been shown for range images but

not for data from pro le scanners as far as we know. A

range image scanner is similar to a pro le scanner but

is not moved relative to the object. The scanner itself

Fig. 2 A laser pro le scanner mounted on an industrial robot

moves the line over the surface of the object, usually by

with a turntable

rotating the laser source around a xed axis. The range

camera thus creates a series of pro les, each related to a

speci c angle of rotation but all in the same coordinate

is described in [20]. A motorized rotary table with two

system.

degrees of freedom and a laser scanner mounted on a

This paper includes a literature review covering laser

computer numerical control (CNC) machine with four

scanning and segmentation. In Section 3, the implemen-

degrees of freedom is described in [33]. Callieri et al.

tations of two segmentation algorithms under consider-

[3] present a system based on a range laser scanner

ation are presented. Section 4 presents the results of

mounted on the arm of an industrial robot in combina-

three different experiments. Finally, in Section 5, we

tion with a turntable. For details on view planning and

conclude the paper and propose future work.

automated 3D object reconstruction and inspection, see

[31].

2 Literature review

2.2 Segmentation

2.1 Laser scanning

Segmentation is a wide and complex task, both in terms

of problem formulation and solution approach in differ-

Laser scanning represents a wide range of related tech-

ent applications. Approaches described in the literature

nologies. In [2], Blais presents a review of 3D digitizing

are usually classi ed in one of the following categories.

techniques with a focus on commercial systems. Pito

and Bajcsy [25] present a simple system by combining a

xed range camera with a turntable. Two more exible 2.2.1 Edge-based approaches

systems are described in [4, 22], where a CMM is used

in combination with a laser pro le scanner. A laser Edge-based approaches attempt to detect discontinu-

pro le scanner with two charge-coupled device (CCD) ities in the surface represented by the point data.

cameras mounted on a three-axis transport mechanism Fan et al. [8] used local surface curvature properties

Int J Adv Manuf Technol (2009) 45:181 190 183

Jain [12] segmented range images into surface patches

Capture profile

and classi ed them as planar, convex, or concave based

data

on a nonparametric statistical test. Besl and Jain [1]

developed a segmentation method based on variable

order surface tting. A region-growing algorithm based

Split profiles and fit on numerical curvature estimation of mesh triangles

line segments was published by Sacchi et al. [28, 29]. Miguel and

Shimada [35] automatically segmented a dense mesh

into regions approximated by single surfaces. The al-

gorithm iterates between region growing and surface

Find the longest line

tting to maximize the number of connected vertices

and use its neighbours

approximated by a single surface. Rabbani et al. [26]

to define a seed plane

segmented a point cloud by using local surface normals

and point connectivity.

Grow the region

2.2.3 Hybrid approaches

around the seed

Hybrid segmentation approaches have been developed

No Seed Found

where the edge-based and region growing-based ap-

Find Next Seed

proaches are combined. The approach proposed by

Yokoya and Levine in [38] divided a point cloud into

surface primitives using biquadratic surface tting. The

Remove over

segmented data were homogeneous in differential geo-

segmented planes

metric properties and did not contain discontinuities.

The Gaussian and mean curvatures were computed

and used to perform the initial region-based segmen-

tation. Then, after employing two additional edge-

Fit planes

based segmentations from the partial derivatives and

depth values, the nal segmentation was applied to

Fig. 3 Main steps of planar segmentation using 2D pro les

the initially segmented data. Checchin et al. [5] used a

hybrid approach that combined edge detection based

to identify signi cant boundaries in range image data.

Chen and Liu [6] segmented data from a CMM by

slicing and tting them with 2D NURBS curves. The

boundary points were detected by calculating the max-

imum curvature of the tted curve. Milroy et al. [23]

used a semiautomatic edge-based approach for orthog-

onal cross section (OCS) models. Yang and Lee in

[37] identify edge points as the curvature extremes

by estimating the surface curvature. Sappa and Devy

[30] propose an algorithm that very quickly processes

large-range images. The proposed algorithm consists

of two steps. First, a binary edge map is generated.

Then, a contour detection strategy is responsible for

the extraction of the different boundaries. Demarsin

et al. [7] presented an algorithm to extract closed sharp

feature lines, which is necessary to create a closed curve

network.

2.2.2 Region growing-based approaches

Approaches based on region growing use local surface

properties to detect continuous surfaces. Hoffman and Fig. 4 Photograph of object 1

184 Int J Adv Manuf Technol (2009) 45:181 190

on the surface normals and region growing to merge

over segmented regions. Zhao et al. [39] employed a

method based on triangulation and region grouping

that uses edges, critical points, and surface normals.

Gotardo et al. [11] used and improved an estimator

in an iterative process to extract planar and quadric

surfaces from range images. Additionally, a genetic

algorithm was speci cally designed to accelerate the

process of surface extraction. Finally, general overviews

and surveys of segmentation methods are provided by

[1] and [24, 32, 36]. In addition, Hoover et al. [13]

present a comprehensive experimental comparison of

techniques for range image segmentation into planar

patches. Fig. 6 Object 1 after planar segmentation using method 1

source or the camera. Since 2D operations are usually

3 Methodology

faster than 3D ones, we want to investigate if P j,i can

be used to speed up computations compared to existing

3.1 Problem formulation

methods operating on S, where all data are 3D. We

will show that this is possible and also compare the

An organized point cloud S is de ned as a set of points

new algorithm with an existing method based on point

in 3D spatially sorted in a topologically triangular or

clouds in three experiments.

rectangular grid. Planar segmentation of S is the par-

titioning of S into planar regions { R1, R2, R3, Rn },

where n is the number of planar regions in S, and 3.2 Planar segmentation using 2D pro les

Ri R j =, i = j, and R S. See Section 2 for ref-

erences to such methods. This paper is concerned with Planar segmentation using 2D pro les is based on the

planar segmentation of data from a laser pro le scan- fact that the image of a straight line projected on a

ner. If such a scanner is moved along a path, it will out- planar surface is also a straight line. Similar algorithms

put a sequence of M pro les F j, where j = 1, 2, 3 M. applied to range images are described in [14, 16]. A

Each pro le F j is de ned in its own coordinate system recursive splitting and line tting algorithm, based on

C j, in 3D, as a sequence of 2D points P j,i = (x j,i, y j,i ), scalar thresholds Dmax and Lmin, is therefore applied as

where i = 1, 2, 3 N j and x j,i 1 Dmax, split the pro le at P j,dmax and apply the test

recursively on the new point sets until all D

Contact this candidate