Page 41 - Autonomous Mobile Robots

24 Autonomous Mobile Robots

definition of obstacle is any object that can obstruct the vehicle’s driving path
or, in other words, anything rising out significantly from the road surface.
Many approaches for extracting obstacles from range images have been
proposed. Most approaches use either a global or a local reference plane to
detect positive (above the reference plane) or negative (below the reference
plane) obstacles. It is also possible to use salient points detected by an elevation
differential method to identify obstacle regions [31]. The fastest of obstacle
detection algorithms, range differencing, simply subtract the range image of
an actual scene from the expected range image of a horizontal plane (global
reference plane). While rapid, this technique is not very robust, since mild
slopes will result in false indications of obstacles. So far the most frequently
used and most reliable solutions are based on comparison of 3D data with
local reference planes. Thorpe et al. [22] analyzed scanning laser range data
and constructed a surface property map represented in a Cartesian coordinate
system viewed from above, which yielded the surface type of each point and its
geometric parameters for segmentation of the scene map into traversable and
obstacle regions. The procedure includes the following.

Preprocessing. The input from a 2D ladar may contain unreliable range data
resulting from surfaces such as water or glossy pigment, as well as the mixed
points at the edge of an object. Filtering is needed to remove these undesirable
jumps in range. After that, the range data are transformed from angular to
Cartesian (x-y-z) coordinates.
Feature extraction and clustering. Surface normals are calculated from x-y-z
points. Normals are clustered into patches with similar normal orientations.
Region growth is used to expand the patches until the fitting error is larger than
a given threshold. The smoothness of a patch is evaluated by fitting a surface
(plane or quadric).
Defect detection. Flat, traversable surfaces will have vertical surface nor-
mals. Obstacles will have surface patches with normals pointed in other
directions.
Defect analysis. A simple obstacle map is not sufficient for obstacle ana-
lysis. For greater accuracy, a sequence of images corresponding to overlapping
terrain is combined in an extended obstacle map. The analysis software can
also incorporate color or curvature information into the obstacle map.
Extended obstacle map output. The obstacle map with a header (indic-
ating map size, resolution, etc.) and a square, 2D array of cells (indicating
traversability) are generated for the planner.

1.3.5.2 Stereo vision
Humans exploit various physiological and psychological depth cues. Stereo
cameras are built to mimic one of the ways in which the human visual system

© 2006 by Taylor & Francis Group, LLC

FRANKL: “dk6033_c001” — 2006/3/31 — 16:42 — page 24 — #24