Traffic Video Surveillance in

Different Weather Conditions

Igor Vujovića, Marjan Jurčevićb, Ivica Kuzmanića

Various parameters are studied in video-stream traffic 1. INTRODUCTION

surveillance. Weather conditions are one of the most important
parameters, influencing the quality of traffic surveillance in any Traffic surveillance is a fast growing area of research.
microlocation, such as the town of Metković in this study. This There are many methods, which address different problems
paper examines the influence of weather conditions on traffic in traffic surveillance. Traffic surveillance can be performed by
surveillance based on the collection of visual data. The quality of different sensors in different situations and types of traffic, from
the system was tested under the following weather conditions: air traffic separation (Williams et al., 2008) to marine integrated
sunny, windy, cloudy and rainy. surveillance (Moutray and Ponsford, 1997).
Cutting-edge studies of traffic surveillance deal with
KEY WORDS
components of intelligent surveillance systems and high-level
~ Vehicle detection
modules of computer vision. However, advances have also been
~ Movement recording
made in low-level processing. An example of both courses of
~ Road traffic surveillance
research is provided in (Cucchiara et. al., 2000), where rule-based
~ Weather conditions
reasoning on visual data is used. The proposed system consists
of several modules, which can be grouped into low-level and
high-level. At high level, symbolic reasoning is introduced. At
low level, two groups of modules are used: for daylight and
nightlight. During the day, motion detection must be corrected
with luminance variation detection. At night, morphology
and headlight pairing is performed. Low-level problems were
addressed i.e. in (Bhandarkar and Luo, 2005), where background
updating is addressed or in (Chang, 2006), where shadow removal
algorithm is proposed. High level problems were addressed i.e. in
(Liu and Payeur, 2003), where activity detection was considered
or in (Kumar et al., 2005), where behavior interpretation was
considered. A method for vehicle count in the presence of
occlusions was proposed in (Pang et al., 2007).
In this paper, we present road traffic surveillance
experiments performed in the Metković region known for
having weather conditions specific only to that region, making
a. University of Split, Faculty of Maritime Studies, Split, Croatia it reasonable and mandatory to examine and investigate
e-mail: [email protected] the impact of weather conditions on the accuracy of traffic
surveillance systems.
b. Zrinsko-Frankopanska 82, Metković, Croatia The paper is organized as follows. In the second section,
e-mail: [email protected] a review of some recent research in traffic surveillance taking

32 doi: 10.7225/toms.v03.n01.004 Igor Vujović et al.: Traffic Video Surveillance in Different Weather Conditions
weather conditions into account is presented. The third section problems are due to the fact that there are no standard camera
describes the setups and the experiments performed in this positions, no standard roads, no standard weather and no
work. The forth section gives conclusions and proposes further standard illumination. It is also stated that different applications
research. require different cameras and different camera positions. It has to
be pointed out that the main problem in video traffic surveillance
2. TRAFFIC SURVEILLANCE UNDER DIFFERENT is the validity of the obtained traffic data, inevitably leading to
WEATHER CONDITIONS the conclusion that weather conditions having to do with the
cleanliness of the camera lens are the main issue. The influence
There are references dealing not only with traffic of fog depends on the camera’s position and density of the fog. If
surveillance, but also including considerations about weather the camera is mounted above the highway and visibility is about
conditions. In this section, references are grouped as: 20 m, than it will have no impact on the results of surveillance.
t
examples of usage from practice (government and private), On the other hand, rain and snow can influence work by making
t
highway (freeway) surveillance, and camera dirty if left unprotected from such conditions. However,
t
urban traffic surveillance. right positioning and protection will reduce the effect of snow
and rain. Also, it goes without saying that the same goes for dirt.
2.1 Government and private sector efforts The camera should be mounted at a sufficiently high position.

The Norwegian Public Roads Administration (NPRA) 2.2 Surveillance of highways

manages the country’s roads for everything from weather
conditions to accidents (Norwegian Roads Agency Deploys The recognition of importance of weather conditions is
Hardened Wireless Network, 2012). This is important for traffic outlined with research in (Lagorio et al., 2008). An analysis of
safety and emergency interventions, as well as for traffic pattern different weather conditions is presented and solution proposed.
research relevant for planning further road network development. The solution is based on the developed system, which uses a
There are also corporations offering complete traffic statistical framework (mixture of Gaussians, MoGs) to identify
surveillance solutions, including weather conditions, which specific meteorological events, such as snow, fog or heavy rain. Of
are intelligent and cover motorways, tunnels and waterways course, the above mentioned examples are used for surveillance
(Intelligent solutions for motorways, tunnels and waterways, of state and interstate roads, not for urban traffic.
2010). (Nookola, 2006) introduced methodology for analyzing
The Arizona Department of Transportation spends about the impact of weather on traffic. The author used correlation
$50,000 per location for the installation of Freeway Management coefficient analysis to establish which weather parameter affects
System (FMS) detection, and another $1,000 per annum for the traffic and daily traffic volume variability under different weather
maintenance of each site. They used a list of starting points to conditions.
establish the common conditions that need to be considered, (Arth et al., 2006) proposed an algorithm that performs
such as: morning peak, afternoon peak, off-peak, dry weather, reasonably well under adverse weather conditions, such as heavy
wet weather, congested conditions with slow speeds, free-flow rain or snow. Errors, however, occur due to shadows or occlusions.
conditions, intense fog, blowing dust, full sunlight, full dark, light Furthermore, the algorithm is sensitive to camera motion, such
transitions, and snow. as jitter. A further drawback is that objects, which become
The (State of the Art Evaluation of Traffic Detection and stationary, tend to fade into the background model. They used
Monitoring Systems, 2007) reports that the quantity of errors the approximated median filter for background modeling and
increases in wet weather. For example, while detected errors in Viola-Jones detector (Viola et al., 2001).
dry weather do not exceed 20 %, they may rise up to as many as (Roh et al., 2013) proved that change in weather conditions
65 % in case of wet weather. The importance of these studies can triggers variations in highway traffic. The goal of the research
be illustrated by investment in experimental setups. was to use truck traffic as a model to plan efficient winter road
Governments have also shown concern for weather maintenance programs.
conditions in traffic monitoring, and ordered the relevant studies. All the above studies indicate that the study of the influence
For example, the US Federal Highway Administration obtained of weather conditions on traffic has practical applications.
results from the Road weather management program (Goodwin,
2003). The report illustrates best practices with examples from US 2.3 Urban traffic surveillance
roads.
An example of European private corporation study may be Dealing with urban traffic opens a variety of issues. Traffic
found in (Versavel, 2006). The report states that traffic surveillance congestions, common in urban areas and their vicinity, are

Trans. marit. sci. 2014; 01: 32-41 TRANSACTIONS ON MARITIME SCIENCE 33

 managed by the intelligent transport system (ITS). ITS integrates is used for that purpose. It depends on buffer length, n, and is
traffic radars, loop detectors and surveillance cameras. given as:
Two main difficulties in the establishment of the virtual
loop detector (VLD) are changing illumination and weather
Bt ( x, y ) = F ( It-1 ( x, y ),..., It-n ( x, y ) ) (1)
conditions. Proposed solution in (Zhidong et al., 2008) increases
average accuracy by 95 % under changing illumination in
various weather conditions in daytime in comparison to relevant where Bt is the background at time t, It(x,y) is the frame coming
previous research. Texture features in describing the traffic from the camera at time t and F is a function that is based on
density are used in (Wassantachat, 2009). Background modeling the temporal variation of pixel values. Function F is obtained
is performed by the SVM kernel, which self-adapts to various in a variety of ways: such as frame differencing, median filter,
lighting environments. linear predictive filter or some non-parametric model. The
Urban traffic is more challenging due to: simplest model is frame differencing, which leads to background
t
high traffic density, definition as:
t
position of cameras, i.e. lower angles resulting in high
occlusion rates, and
Bt ( x, y ) = It-1 ( x, y ) (2)
t
variety of road users.
Furthermore, evaluation under challenging weather
conditions would be desirable, but is rarely performed (Buch et
This, the simplest method, is only based on the previous
al., 2011). Different techniques for urban road user detection and
frame. The calculation of the median for several frames in the
classification are proposed, such as in (Buch et al., 2010), where
buffer is a bit more complex:
motion silhouettes were extracted and compared to projected
model silhouettes. The comparison resolved the issue of the
classification of motion silhouettes. Performances were tested
Bt ( x, y ) = median ( It-1 ( x, y ),..., It-n ( x, y ) ) (3)
under three different weather conditions and results were 87 %
better than in competitive algorithms.
One of the challenging problems in urban traffic is In linear predictive filters, background model is defined as:
the tracking of vehicles at crossroads. In (Liwei at al., 2011),
a new method is proposed. The method is innovative in the n
application of two-stage view selection and dual-layer handling Bt ( x, y ) = ∑ αi It-i ( x, y )i (4)
of occlusions. The paper proposed a Multi-Modal Particle Filter i=1

(MMPF) for tracking vehicles explicitly. A considerable number of

experiments were performed under different weather conditions. where αi is the predictive coefficient.
The considered weather conditions were snowy, sunny and The most popular recursive techniques mentioned are the
cloudy. approximated median, Kalman and mixture of the Gaussians.
(Smids, 2006) used webcams (in our example, webcam was (Smids, 2006) used Open Source Computer Vision Library
also used) for urban traffic monitoring applying the background (OpenCV) for the implementation of the proposed work. Finally,
subtraction technique. The proposed system was evaluated (Smids, 2006) concluded that regardless of the weather conditions,
under different weather conditions. The author compared the statistical approach outperforms the deterministic approach
the deterministic and statistical approach in background in video surveillance applications. However, he reported a
subtraction and background modeling. The author reported number of small misclassification intervals due to rainy weather.
that deterministic subtraction approach had some important (Zhou et al., 2007) reported, according to their experimental
limitations. For example, total foreground area was often not fully results based on real traffic video data, that their vehicle detector
detected, objects which became background were erroneously was excellent at dealing with different weather and illumination
detected as foreground. However, results based on statistical conditions. They based their conclusion on counted vehicles
approach using per-pixel Gaussian mixture model (GMM) were passing by the area under different weather and illumination
reported as very promising. The deterministic approach has no conditions: afternoon sunlight with long and heavy shadow,
limitations. Sunny, cloudy and rainy weather conditions were morning sunlight with long and light shadow, a rainy day with
used in the research. Experimental work was performed with reflections on the side of vehicles, nightfall with bright vehicle
a capture resolution of 640x480 and frame rate was 15 fps. The headlights but no streetlight and at night under streetlight with
author is of the opinion that background should be adaptive. In shadows and headlights. Their approach was the implementation
non-recursive modeling techniques, a sliding window approach of SVM (Support Vector Machine) in training. They obtained

34 doi: 10.7225/toms.v03.n01.004 Igor Vujović et al.: Traffic Video Surveillance in Different Weather Conditions
SVM-classifiers, which they used in further experimental work. (Narasimhan and Nayar, 2003) concluded that there is no
The scatter matrix for PCA (Principal Component Analysis) is point in improving classic noise-removal techniques to remove
computed using a training sequence. Finally, they concluded the influence of weather from the image. The reason is in contrast
that the experimental results derived from real traffic video data decay, which is caused by weather. They developed a physics-
showed that their vehicle detector has an ability to deal with based model which describes the appearances of scenes in
different weather conditions. uniform bad weather conditions.
As stated in (Nandhini and Parthiban, 2012), a video-based As can be seen, some papers concentrate on low level
monitoring system must be resistant to different weather and and some on high level processing of the input video stream.
illumination conditions. The algorithm deals with shadows as However, both low and high level processing have the potential
well. The proposed algorithm includes lighting object extraction, for further research due to the above mentioned issues, which
connected-component extraction, spatial classification process, still remain unresolved/unresearched or not resolved in a
potential component tracking, motion-based grouping, tracking satisfactory manner.
of vehicle component groups and vehicle identification and
classification. In the research, satellite images were used. The 3. EXPERIMENTAL SETUP AND RESULTS
preprocessing of such images consists of several operations, such
as atmospheric correction, change detection, image registration In this section, we described the performed experimental
or geometric correction. Vehicle lights were used as the silent setup, results, and measures of quality. As previously stated,
feature for nighttime vehicle detection. Then morphology the Metković region is known for having weather conditions
was used, as well as shape analysis, etc. Edge-based detection specific only to that region. So it is reasonable and mandatory
experiments were performed under different weather and to research and investigate the impact of weather conditions on
lighting conditions. Finally, the authors concluded that their the accuracy of the traffic surveillance systems. The quality of the
proposed algorithm can be used in real-world and real-time road system is examined for the following weather conditions: sunny,
traffic applications. windy and cloudy, and rainy, since these weather conditions are
(Cheung and Kamath, 2005) are more concerned with the the most frequent in the geographical microregion of the town
operation of the standard modules, which every reliable video of Metković, Croatia.
surveillance system of differencing type must have – robust
background subtraction. They validated their proposal for 3.1 Description of experimental setup
background subtraction in urban traffic video.
(Hu et al., 2008) focused on data fusion provided by Equipment used in the experiments is as follows: network
different viewpoints and the probability of vehicle presence. camera AXIS 207, software surveillance application, laptop
They submitted that noise caused by weather, sudden light HP Compaq 6715b. Characteristics of the camera Axis 207 are
changes and weak shadows could be efficiently eliminated using as follows: dimensions 85x55x34 mm, weight 177 g, network
their method. The authors used a concept of probability fusion connection RJ45, grid adapter 4 W power, and the input/
map (PFM), which is the probability of a vehicle being present in output terminal connector (Axis 207/207W/207MW Installation
the scene. PFM is calculated by merging mapped images from Guide, 2012). MJPG format was used for processing. Image
different cameras. The additional benefit of using more than one resolution was 640x480 pixels. Network cameras make possible
camera is the ability to obtain 3D information, such as length, the organization of traffic surveillance in control centers. Such
width and height in the process of blob analysis. They proposed cameras are organized as free-standing and can be contacted by
a key factor called PFM factor, which synthesizes the inverse different authorized computers. The advantage is the possibility
projected images with considerations of Inverse Projector Factors of networking many cameras into a single system.
(IPF) and Perspective Accuracy Factor (PAF). The IPF depends on Further advantages of the cameras used are (Axis net
the properties of the pixel and PAF is expressed by factor β: cameras, 2012):
t
remote and safe access to live video stream and recorded
videos,
Pi ( x, y ) t
flexible and economic solutions to video control,
βi ( x, y ) = 1 κi log10 (5) t
more reliable identification,
Pi0
t
PAL/NTSC may be used, but even higher resolutions can be
achieved,
where Pi0 is camera distance to the nearest feature point, κi an t
progressive scanning to remove blur,
adjustment weight parameter and Pi (x,y) is the distance from t
powerful event management with intelligent video
target to camera calculated on the inverse projected map. features,

Trans. marit. sci. 2014; 01: 32-41 TRANSACTIONS ON MARITIME SCIENCE 35

 t
automatic reaction to events and threats, drastically 3.2 Results
reducing the number of operative personnel and raising the
efficiency of video surveillance, Camera was positioned at state road D62 in the vicinity of
t
scalable, easily integrated solution, the town of Metković in the Republic of Croatia.
t
open access technology platform, The first experiment was performed under sunny weather
t
easy integration into other systems as access control or conditions. As shown in Section 2, weather conditions influence
point of sale. the performance of the systems. Therefore, different weather
t
lower total costs, conditions have been experimentally studied. The examples of
t
power over ethernet is possible, reducing the costs of the results are shown in Figure 1.
installation at remote locations, and
t
connection to standard IP network/ethernet.

a) c)

b) d)

Figure 1.
Motion detection in experiment 1(camera is not directly exposed to the sun): a) no motion, b) c) d) motion detected.

In the experiment, the camera was tested in two modes The second experiment was performed in windy and cloudy
of operation: the shooting of moving images (when there is no weather. Generally known problems occurring under these
motion, no picture is not taken) and constant recording of traffic. weather conditions are: waving trees, thresholding, detection of
Constant recording is useful if motion detector fails to trigger cloud movement and so on. In some cases, some of the problems
the recording device. It could be examined how well the motion can be avoided by adding specific program routines. This solution
detector and hardware operate in real conditions. Both software cannot be used generally.
and hardware support both modes of operation.

36 doi: 10.7225/toms.v03.n01.004 Igor Vujović et al.: Traffic Video Surveillance in Different Weather Conditions
 a) c)

b) d)

Figure 2.
Motion detection under windy and cloudily weather: a) no motions, b), c), d) wrong detection of motions: waving trees effect.

The third experiment was performed in rainy conditions. In the camera does not have features which would allow it to
this case, detection of rain as motion presents a problem, which differentiate between heavy rain and movement. Therefore,
could be avoided by setting a proper threshold. Unfortunately, there is no rain sensibility.

a) c)

Trans. marit. sci. 2014; 01: 32-41 TRANSACTIONS ON MARITIME SCIENCE 37

 b) d)

Figure 3.
Motion detection during rain: a) rain as detected motion, b), c), d) real motion detection.

In order for there to be a noticeable difference, frame technique. This type of image is called a motion mask. It is clear
differencing has to be performed. The used application software from the motion mask that shadow detection can be a very real
enables frame differencing. The result is in Figure 4. Figures 4.a problem when we want to use the obtained motion mask further
and 4.b show two neighboring frames in the video stream. Figure in high level applications. The second problem of the application
4.c presents the difference between Figures 4.a and 4.b. It can be software used is the avoidance of double detection. A higher
seen that frame differencing can produce double-vision effect vision application, like car counting, can instigate false detection
in the motion mask. The result is presented in black and white by mistake.

a) b) c)

Figure 4.
Frame differencing: a) frame 1, b) frame 2, c) motion mask obtained by frame differencing.

The result of image sequence processing is the image applications, i.e. traffic counting, which will enable traffic
difference. Further research should include higher vision development planning.

38 doi: 10.7225/toms.v03.n01.004 Igor Vujović et al.: Traffic Video Surveillance in Different Weather Conditions
3.4 Measures of quality False detection rate can be defined as:

Measures of quality can be divided into 11 groups based

FP
on: FDR = (9)
t
pixels, TP + FP
t
objects,
t
spatial precision, In some cases, an interesting measure can be false negative
t
temporal and spatio-temporal precision, detection rate defined as:
t
estimation methodology,
t
receiver-operating characteristics,
FN
t
perturbation detection rate, FNDR = (10)
t
standalone performance evaluation, TP + TN + FN + FP
t
intra-object homogeneity metrics,
t
inter-object disparity metrics, and The total number of false detections, total false detection
t
combination of the above mentioned. rate, TFDR, is defined as:
In pixel based quality measures, pixels are classified into
four groups: FN + FP
t
true positives (TP), which means that the algorithm TFDR = (11)
correctly detected and classified the pixel as foreground, TP + TN + FP + FN
t
false positives (FP), which means that the algorithm
mistakenly classified a pixel as foreground instead of as The issue of statistical pixel measurement is possibly
background, misleading. It is especially noticeable in scenes containing
t
true negatives (TN), which means that the algorithm changes small relative to the total number of pixels.
correctly classified a pixel as background, Although misleading in many situations, the pixel criterion
t
false negatives (FN), which means that the algorithm is easy to implement, does not require high vision algorithms and
classified a pixel as background instead of as foreground. high execution times.
Based on the above mentioned definitions, several For example, if we take Figure 4, we can measure that PCC
measures are introduced. The best known is the percentage of is 0.74, which is lower than many methods in video surveillance.
correct classifications (PCC) (Rosin and Ioannidis, 2003; Elhabian However, since all of these methods should be evaluated under
et al., 2008): the same conditions, using the same scenes, this is nothing more
than an assumption, not a scientific fact.
TP + TN Further numerical result is the FNDR, which is only 0.021,
PCC = (6) because we did not take shadows and double-detections into
TP + TN + FP + FN consideration. However, when shadows and double detections
are taken into account, the appropriate measure, FDR, rises to
The second measure based on pixels is the so called Jaccard 0.2381 or 23.8 %, which is a very bad result in comparison to the
coefficient (JC), defined as: world’s best results. TFDR is almost 26 %. One of the problems
with real scenes and the evaluation of new experiments is the lack
of grounds for comparison. There are not many publicly available
TP and compatible source codes to compare with your scene. On
JC = (7)
the other hand, if the available reference video sequences are
TP + FP + FN
used, one risks being misled, because there is no guarantee that
the obtained results can be applied to all scene in general.
The third statistical measure is Yule’s coefficient (YC): Unfortunately, the research results from available literature
focus on higher vision applications, such as traffic patterns, time
predictions, speed predictions, etc., i.e. in (Roh et al., 2013). For
TP TN
YC = + -1 (8) example, vehicle counting is influenced by weather conditions,
TP + FP TN + FN i.e. in (Zhou et al., 2007) rainy conditions contribute to false car
counting by less than 1 %. Error rates increase by minimum
20% in humid weather (see State of the Art Evaluation of Traffic
Detection and Monitoring Systems, 2007).

Trans. marit. sci. 2014; 01: 32-41 TRANSACTIONS ON MARITIME SCIENCE 39

 4. CONCLUSION http://dx.doi.org/10.1109/TITS.2011.2119372

Chang, F. L., (2006), Object Tracking, Shadow Removal and Collision Event Detection
The study of the influence of weather conditions was for Traffic Surveillance System, MSc Thesis, National Cheng Kung University, Taiwan.
justified due to the specific climate characteristics of the Cheung, S. C. S. and Kamath, C., (2005), Robust Background
Metković region. The goal of this study was to improve automatic Subtraction with Foreground Validation for Urban Traffic Video,
traffic monitoring and control systems. The following weather EURASIP Journal on Applied Signal Processing, 14, pp. 2330–2340.,
condition were studied: sunny, rainy, and cloudy weather. The http://dx.doi.org/10.1155/ASP.2005.2330

obtained results lead to the conclusion that the choice of the Cucchiara, R., Piccardi, M. and Mello, P., (2000), Image Analysis
selection of the measure for the decision making process is fuzzy. and Rule-Based Reasoning for a Traffic Monitoring System, IEEE
A smart and intelligent system, containing event bases, must be Transactions on Intelligent Transportation Systems, 1(2), pp. 119-130.,
http://dx.doi.org/10.1109/6979.880969
included into research and development.
Experiments performed in the study revealed possible issues Elhabian, S. Y., El-Sayed, K. M. and Ahmed, S. H., (2008), Moving Object
Detection in Spatial Domain using Background Removal Techniques
when dealing with outdoor applications of video surveillance,
– State-of-Art, Recent Patents on Computer Science, 1(1), pp. 32-54.,
such as wrong detection of clouds as road movement, rain as http://dx.doi.org/10.2174/10032
motion or the waving tree effect.
Goodwin, L. C., (2003), Best Practices for Road Weather Management, Report,
The applications of road traffic surveillance are the study
Road Weather Management Program, Office of Transportation Operations, Federal
of drivers’ behavior, traffic regulation violations, accident Highway Administration.
detection, traffic regulation, security, safety, and similar. Any of
Hu, Z., Wang, C. and Uchimura, K., (2008), 3D Vehicle Extraction and Tracking
these accidents can be reported to the police or a monitoring
from Multiple Viewpoints for Traffic Monitoring by using Probability Fusion Map,
station, and further actions can be taken under advisement. In Electronic Letters on Computer Vision and Image Analysis, 7(2), pp.110-119.
the research, motion mask was the basis for the calculation of the
Intelligent Solutions for Motorways, Tunnels and Waterways, Siemens,
measures of quality.
(2010), available at: http://www.mobility.siemens.com/mobility/global/
The experiments were performed as a part of requirements SiteCollectionDocuments/en/road-solutions/interurban/intelligent-solutions-
for a master’s thesis (MSc). Further work could lead to the motorways-tunnels-waterways-en.pdf, [accessed 14 November 2012.].
framework for the assessment of the influence of weather on
Kumar, P., Ranganath, S., Weimin, H. and Sengupa, K., (2005), Framework
traffic monitoring based on low-level image processing criteria. for Real-time Behavior Interpretation from Traffic Video, IEEE
One of the future papers could deal with the calculation of the Transactions on Intelligent Transportation Systems, 6(1), pp. 43-53.,
velocity of a moving object based on frame differencing. Possible http://dx.doi.org/10.1109/TITS.2004.838219
issues could include slight camera vibrations or variations in Lagorio, A., Grosso, E. and Tistarelli, M., (2008), Automatic Detection of Adverse
position, which would result in an error in calculation. Small Weather Conditions in Traffic Scenes, Proc. IEEE Fifth International Conference on
displacements can be caused by natural factors, i.e. the wind. Advanced Video and Signal Based Surveillance AVSS ‘08, Santa Fe, USA, September
1-3, Curran Associates, pp. 273 – 279.

REFERENCES Li, Z., Tan, E., Chen, J. and Wassantachat, T., (2008), On Traffic Density Estimation with
a Boosted SVM Classifier, Digital Image Computing: Techniques and Applications
Arth, C., Bischof, H., Leistner, C., (2006), TRICam – An Embedded Platform for (DICTA), Canberra, Australia, December 1-3, IEEE Computer Society, pp. 117-123.
Remote Traffic Surveillance, Proc. 2006 Conference on Computer Vision and
Pattern Recognition Workshop (CVPRW’06), New York, June 17 – 22, pp. 125., Liu, L., Xing, J. and Haizhou, A., (2011), Multi-view Vehicle Detection and Tracking
http://dx.doi.org/10.1109/CVPRW.2006.208 in Crossroads, Proc. 2011 First Asian Conference on Pattern Recognition (ACPR),
Beijing, China, November 28-28, pp. 608 – 612.
Axis 207/207W/207MW Installation Guide, available at: http://www.axis.com/files/
manuals/27192.pdf, [accessed 5 October 2012.]. Liu, Y. and Payeur, P., (2003), Vision-based Detection of Activity for
Traffic Control, Proc. IEEE Canadian Conference on Electrical and
Axis net cameras, available at: http://www.axis.com/files/brochure/bc_ Computing Engineering, Montréal,Canada, 2, May 4-7, pp. 1347 – 1350.,
netcams_40664_hr_1009_lo.pdf, [accessed 2 October 2012.]. http://dx.doi.org/10.1109/CCECE.2003.1226149
Bhandarkar S. M. and Luo, X., (2005), Fast and Robust Background Updating for Real- Moutray, R.E. and Ponsford, A.M., (1997), Integrated Maritime Surveillance (IMS) for
time Traffic Surveillance and Monitoring, Proceedings of the 2005 IEEE Computer the Grand Banks, MTS/IEEE Conference Proceedings OCEANS ‘97, October 6-9, 2, pp.
Society Conference on Computer Vision and Pattern Recognition CVPR’05, San 981 – 986.
Diego, USA, June 25-25, IEEE Computer Society, pp. 55 – 55.
Nandhini, S. and Parthiban, P.G.S., (2012), Automatic Vehicle Detection during
Buch, N., Orwell, J., and Velastin, S. A., (2010), Urban Road User Detection and Nighttime Using Bright Pixel Segment with Spatial Temporal Technique,
Classification Using 3D Wire Frame Models, IET Computer Vision, 4(2), pp. 105 – 116., International Journal of Electronics and Computer Science Engineering, 1(3), pp.
http://dx.doi.org/10.1049/iet-cvi.2008.0089 1520-1525.
Buch, N., Velastin, S. A. and Orwell, J., (2011), A Review of Computer Narasimhan, S. G. and Nayar, S. K., (2003), Contrast Restoration
Vision Techniques for the Analysis of Urban Traffic, IEEE Transactions of Weather Degraded Images, IEEE Transactions on Pattern
on Intelligent Transportation Systems, 12(3), pp. 920 – 939.,

40 doi: 10.7225/toms.v03.n01.004 Igor Vujović et al.: Traffic Video Surveillance in Different Weather Conditions
Analysis and Machine Intelligence, 25(6), pp.713-724., UTP cable for LAN, available at: http://www.ic.ims.hr/faq/mreza/utp/utp.html
http://dx.doi.org/10.1109/TPAMI.2003.1201821 [accessed 6 October 2012.].

Nookola, L. S., (2006), Weather Impact on Traffic Conditions and Travel Time Zhou, J., Guo, D. and Zhang, D., (2007), Moving Vehicle Detection for Automatic
Prediction, MSc Thesis, University of Minnesota Duluth. Traffic Monitoring, IEEE Transactions on Vehicular Technology, 56(1), pp. 51-59.,
http://dx.doi.org/10.1109/TVT.2006.883735
Norwegian Roads Agency Deploys Hardened Wireless Network, Cisco Ltd.,
(2012), available at: http://www.cisco.com/en/US/prod/collateral/routers/ps380/ Versavel, J., (2006), Traffic Data Collection: Quality Aspects of Video Detection,
c36_703866_00_nra_cs.pdf, [accessed 14 November 2012.]. Report, Traficon n.v., Bissegem, Belgium, available at: http://sites.poli.usp.br/ptr/
lemt/documents/07-0901.pdf, [accessed 10 October 2013.].
Open Source Computer Vision library, available at: http://www.intel.com/
technology/computing/opencv/, [accessed 14 November 2012.]. Viola, P. A. and Jones, M. J., (2001), Rapid Object Detection using a Boosted Cascade
of Simple Features. In International Conference on Computer Vision and Pattern
Pang, C.C.C., Lam, W.W.L. and Yung, N.H.C., (2007), A Method for Vehicle
Recognition, Kauai, Hawaii, December 11-13, pp. 511–518.
Count in the Presence of Multiple-Vehicle Occlusions in Traffic Images,
IEEE Transactions on Intelligent Transportation Systems, 8(3),pp. 441-459., Wassantachat, T., Li, Z., Chen, J., Wang, Y. and Tan, E., (2009), Traffic Density
http://dx.doi.org/10.1109/TITS.2007.902647 Estimation with On-line SVM Classifier, Proc. Sixth IEEE International Conference on
Advanced Video and Signal Based Surveillance AVSS ‘09, Genova, Italy, September
Roh, H. J., Datla, S. and Sharma, S., (2013), Effect of Snow,
2-4, Conference publishing Service, pp. 13–18.
Temperature and Their Interaction on Highway Truck
Traffic, Journal of Transportation Technologies, 3, pp. 24-38., Williams, J. T., McGraw, J. W., Pate, D. P. and Porter, S., (2008), Reducing Separation
http://dx.doi.org/10.4236/jtts.2013.31003 Requirements through Improved Navigation and Surveillance, Proc. Tyrrhenian
International Workshop on Digital Communications - Enhanced Surveillance of
Rosin, P. and Ioannidis, E., (2003), Evaluation of Global Image Thresholding
Aircraft and Vehicles TIWDC/ESAV 2008, Island of Capri, Italy, September 3-5, pp.
for Change Detection, Pattern Recognition Letters, 24(14), pp. 2345-2356.,
1 – 6.
http://dx.doi.org/10.1016/S0167-8655(03)00060-6

Smids, M., (2006), Background Subtraction for Urban Traffic Monitoring using
Webcams, MSc Thesis, Universiteit van Amsterdam.

State of the Art Evaluation of Traffic Detection and Monitoring Systems, (2007), test
report Texas Transportation Institute for Arizona Department of Transportation,
available at: http://www.azdot.gov/tpd/atrc/publications/project_reports/pdf/
az627(1).pdf, [accessed 15 September 2012.].

Trans. marit. sci. 2014; 01: 32-41 TRANSACTIONS ON MARITIME SCIENCE 41