IDENTIFICATION AND LOCATION OF SIMPLE OBJECTS FOR REAL-TIME
MAPPING
Christoph Geiselmann Michael Hahn
RIB Bausoftware GmbH Institute of Photogrammetry
Vaihinger Str. 151 Stuttgart University
maaan 70567 Stuttgart / Germany P.O.B. 106037
70049 Stuttgart / Germany
mee] ISPRS Commission II , Invited paper
KEY WORDS: Classification, real-time mapping, invariant features, affine-invariance, colour image
sequence, motion based segmentation, traffic signs
EN ABSTRACT
Systems for real-time mapping can be described like most information systems by components for
recording, storage, management, analysis and presentation of the data. The recording and storage of
data collected by positioning and imaging sensors is done in real-time. Up-to-date real-time is not fully
achieved by the other system components. In particular the analysis of the images is often complex
1 and time consuming.
In this paper, an investigation in the identification and location of simple objects is presented. As an
example traffic signs recorded by a colour image sequence are used. As a by-product this study shows
a possibility on how to reduce the user activity considerably in the process of creating traffic sign
maps. The developed automatic procedure consists of the following steps: a motion stereo pair is used
for image segmentation into a region of interest and in background. As a result a precise displacement
—— vector field of this region is derived. The deconvolution of motion blur leads to a restoration of the
image and prepares the image for further processing. Next the borderline of the region is determined
— and affine-invariant features of this contour are determined. For interpretation the suitability and
A separability of the features is of special interest. Finally, maximum likelihood classification with these
—— features leads to the identification of object classes for the localized traffic signs. The reported exper-
iments in this investigation are based on simulations and on real world scenes.
isation
e du
459
