In: Paparoditis N.. Pierrot-Deseilligny M.. Mallet C.. Tournaire O. (Eds), IAPRS. Vol. XXXVIII. Part ЗА - Saint-Mandé, France. September 1-3, 2010 
The paper is organized as follows: first, the proposed method is 
described in Section 2, then, the results of experiments and 
discussions are presented in Section 3, followed by conclusions 
and references. 
2. APPROACH 
2.2.1 Congested traffic 
Numerous investigations (e.g. Greensfield. 1935; Kockelmann, 
1998) on real traffic data show that under congested conditions 
the following assumption is true: a class of vehicle's spacing is 
a linear function of the speed of all vehicles 
Our approach for traffic parameter estimation in sequences of 
optical images consists of a two level method (see Figure 1). 
Si = B, ■ g(v) + L t , 
(1) 
First, traffic classification is performed into three main classes: 
free flow, congested and stopped traffic. Then, for each traffic 
class, modelling of traffic flow on the road segments is 
performed separately allowing the direct derivation of the 
required traffic parameters from the data, such as the vehicle 
density and average speed. Further, other traffic information, 
like the existence of congestion, the beginning and end of 
congestion, the length of congestion, actual travel times, and so 
on can be easily extracted. The proposed method is based on the 
combination of various techniques: change detection, image 
processing and incorporation of a priori information such as 
road network (NAVTEQ), information about vehicles and roads 
and finally traffic models. The change detection in two images 
acquired with a short time lag (~ 2 sec) is implemented using 
the Multivariate Alteration Detection (MAD) method (Nielsen 
2007) resulting in a change image where the moving vehicles 
on the roads are highlighted. Image processing techniques can 
be applied to derive the vehicle density in the binarized and 
denoised change image. This estimated vehicle density can be 
related to the real vehicle density, acquired by modelling the 
traffic flow for a road segment. The model is derived from a 
priori information about the vehicle sizes and road parameters, 
the road network and the spacing between the vehicles 
(Palubinskas 2009, 2010). 
2.1 Traffic classification 
where spacing 5, is the front-to-front vehicle distance in meter, 
B, is a dimensionless parameter of the model, function g(v) 
transforms speed (km/h) into meters, e.g. g(100 km/h) = 100 m, 
L■, is the vehicle length in meter and / is the vehicle class, e.g. 
passenger car or truck. Parameter B can be interpreted in the 
following way: for B=0.5 and L=0 the formula (1) means for all 
drivers a w'ell-known rule of thumb “safe distance = half 
speedometer reading in metres". As already mentioned this 
model w^ell describes a congested traffic. The B value is ranging 
normally between 0.5 and 1.0 (Palubinskas 2009). 
Now the traffic density can be calculated as 
D(#vehicles per km) 
1 km 
(2) 
where S = L Pi ■ S , » Pi is a proportion of vehicle class i and 
/ 
^ Pi = 1. Thus for density calculation in (2) the weighted 
7 
mean value of vehicle spacing is used. 
The modelled vehicle density D is directly related to the 
average vehicle speed on the road segment and thus the 
information about the traffic situation can be derived (see 
Figure 2). Detailed description of the proposed method is 
provided in (Palubinskas 2009). 
Traffic classification is performed in order to group vehicles on 
the road into three main classes: free flow, congested and 
stopped traffic. There exist already some approaches for this 
task based on a sequence of images (e.g. Zeller 2009). We did 
not found that they are superior to our proposed single image 
traffic classification which additionally allow's direct derivation 
of approximate vehicle density and thus average speed for a 
road segment. First, a colour (RGB) image is transformed to a 
gray (intensity) image using YCbCr colour transformation 
(YCbCr 2010), which is believed to represent better the 
intensity of colour image then a simple mean. Then, the image 
is shifted in along road direction by a half of vehicle size and a 
difference of original intensity image and its shifted version is 
calculated (the roads are straightened before using NAVTEQ 
road database). Finally, a sum of absolute differences (SAD) is 
calculated for a road segment of a specified length. Simple 
thresholding technique can be applied to classify SAD values 
into three classes: low values correspond to free flow traffic, 
middle values - congested traffic and high values - stopped 
traffic. Additionally, binarized and denoised SAD image can be 
used to estimate roughly vehicle density d c for a given road 
segment. 
2.2 Traffic models 
For each traffic class a different traffic model is applied in order 
to derive desired traffic parameters. 
2.2.2 Stopped traffic 
For stopped vehicles, the previously introduced way of 
estimating vehicle density is no more valid: the values are 
significantly underestimated or even approaching zero due to 
the subtraction of vehicle blobs in a change image. Of course, 
the same model can be applied if vehicle density d c estimated 
during the traffic classification level is used and thus a speed of 
very slowly moving vehicles can be estimated. 
2.2.3 Free flow traffic 
For free flow traffic a very simple model is used: that is a 
maximal allowed speed for a particular road segment is 
assumed thus again leading to a rough speed estimate. 
3. EXPERIMENTS 
To confirm our idea and to validate the method several flight 
campaigns with the DLR airborne experimental wide angle 
optical 3K digital camera system operated on a Do-228 aircraft 
were performed. In this paper one of these experiments is 
presented. Since the area covered is quite large, the evaluation 
is performed for many road segments and can therefore be 
regarded as representative measures.