June 30

Afstudeercolloquium Jasper Smit

Event information

Afstudeercolloquium Jasper Smit

Date Time

Thu, Jun 30, 2011

Location

Bernoulliborg 165

Organiser

Board

Afstudeercolloquium
Jasper Smit
Donderdag 30 juni 2011
10.00 uur
zaal BB 0165

Road Sign Recogntion Using Binary Partition Trees

Road Sign Recognition systems improve road safety by assisting drivers. For this
project a road sign recognition system was created. The system uses the binary
partition tree for the purpose of segmenting the image into regions. A binary
partition tree is, like the max-tree, a hierarchical, representation of the input
image. A max-tree requires a definition of an ordering on its input values, while the
binary partition tree only requires the distance between pixel values. The binary
partition tree is therefore naturally better suitable for color images. Vector
attributes are used to describe the shapes in the image. To find the objects of
interest, a comparison is made between the attribute description of the shape and a
reference vector taken from prototypes of road signs. The prototypes used can be
created in two ways. They can either be defined manually, or they can be extracted
automatically from annotated example images.

In this talk I will discuss the binary partition tree and its implementation. I also
discuss the related and more efficient salience tree. Then the application of the
binary partition tree to the problem of recognizing road signs is discussed and the
performance of the system is evaluated.