||Detection, Recognition, and Georeferencing of Traffic Signs in Street-Level Imagery
||Henrik, Møller Rasmussen
Johan, Musaeus Bruun
||Aanæs, Henrik (Image Analysis and Computer Graphics, Department of Informatics and Mathematical Modeling, Technical University of Denmark, DTU, DK-2800 Kgs. Lyngby, Denmark)
||Technical University of Denmark, DTU, DK-2800 Kgs. Lyngby, Denmark
||For road equipment maintenance, an automated indexing system is desirable. As
a pilot project for such a system, opportunities for detecting, recognizing and
georeferencing traffic signs in GPS-annotated street-level imagery are investigated.
Five important and frequent sign categories are targeted and a colorand
shape-based approach for detecting these is implemented. For recognizing
traffic signs, i.e. identifying pictograms, experiments with a multilayer perceptron
neural network on normalized images are performed. Finally an intuitive
approach for georeferencing signs using spherical geometry and geographical
data without distance measurements is developed. The detection works very
well on warning, yield, and prohibition signs, and decently on mandatory and
information signs, although the latter group has a higher rate of false positives.
The recognition experiments show that classifying pictograms using multilayer
perceptrons seems possible. Since the amount of data was limited, experiments
with additional data are required in order to provide a more robust evaluation.
The intuitive approach for georeferencing shows great results despite factors
such as car roll and pitch, lens distortion, and object elevation being ignored.
Based on the investigations and experimental results, a fully automated sign
indexing system denitely seems achievable.
||Technical University of Denmark (DTU) : Kgs. Lyngby, Denmark
Creation date: 2010-10-11
Update date: 2010-10-11