7B-2-2 
approach is to evaluate the method i.e. to point out the 
advantages and weakness of the algorithm. 
The automatic measurement of the road widths involves colour 
based step-wise processing i.e. segmentation of the road, road 
boundary extraction, stereo measurement of the road with. The 
colour information is used in segmentation and in road boundary 
detection. The road boundaries are found based on edge 
detection and linking withHough transforms. The road width is 
measured with help of epipolar geometry and cross-correlation in 
colour space. 
2 AUTOMATIC MEASUREMENT OF ROAD WIDTHS 
To measure the road width automatically some questions have to 
be investigated, for example: How to segment road surface from 
the image? What is characteristic for the road? Is the image 
information sufficient to properly extract the road from the other 
object? What is road width? How is it defined? Is it a distance 
between the edges of asphalt surface? Or a distance between 
painted lines at the road edges? Some of the answers are quite 
natural and easy, but only if human experience and interaction 
can be used. 
2.1 Road model 
Road segmentation and extraction of the road boundaries is well 
known in the field of autonomous navigation and computer 
vision based guidance. Typically a model is used to guide 
recognition and to predict how and where the road should appear 
and what methods should be used to find itModels like MART 
(Waxman et al„ 1987), VITS (Turk et al„ 1988), SCARF 
(Crisman et al., 1993), and FERMI (Kluge et al., 1989) were 
developed and implemented into in road following vehicles 
(Yang H., 1995). Those models usually consist of an implicit 
model and a subconscious model. The subconscious model 
includes assumptions like that the road does not change colour, it 
is continuously connected or that it is always in camera view. 
Implicit models represents decisions made by a programmer but 
not available to the program. Typically the implicit models 
includes statements that the road is always different from the 
surrounding, or that locally the road can be consider as a straight, 
or that the dominant edges in the scene are the road borders 
(YangH., 1995). 
In this approach the following implicit model was applied: 
• The road has a specific colour, which differs from the 
surrounding. 
• The road is always in front of the vehicle. 
• The road boundaries create dominant edges in gradient 
image. 
• Detection of road edges can be simplified to straight- 
line detection within a Region of Interest (ROI). 
The road segmentation and boundary detection analysis was 
limited to single image only. The idea was to detect two 2D lines 
representing left and right boundary of the road. Then, to find 
two points of intersection of the boundary lines and a horizontal 
line. Next, to determine the 3D position for each intersection 
points (stereo image analysis) and the 3D vector between those 
points. 
The model applies slightly modified algorithms and C++ classes 
developed during research done on topic of automatic road sign 
detection and automated measurements of 3D points and 3D 
vectors (Gajdamowicz, 1998). 
2.2 Segmentation of the road, extraction of the road 
boundaries and measurement of the road width 
Segmentation of the road is based on colour. The initial colour 
values were measured and applied into an algorithm in order to 
obtain a threshold image Gj. This binary image consists of dense 
and scattered pixels (Fig. 1). Typically, scattered pixels are 
created from features with similar colour, as the road surface e.g., 
trees, bushes, stones, etc., are considered as noise, therefore 
should be removed from the image data. Morphological filtering, 
median filtering and Connected Component Analysis (CCA) 
obtain noise reduction. Moreover, the morphological filtering 
followed by CCA is used to calculate theROIs. The biggest ROI 
is considered as a road, where all the other ones, considered as 
noise, are rejected. 
Figure 1 .Segmentation of road: Colour image, threshold image 
Gj, ROI 
The colour image within the ROI was transformed to grey scale 
images corresponding to hue G//, saturationGs and combined 
hue and intensity image G/+//(Eq. 1) 
>G w ,G 5 ,G /+w (1)