XXIX-B3, 2012
eview and analysis of
r classification of
‘ernational Journal of
H., Shibayama, M.,
phenology detection
Remote Sensing of
new forest cover map
derived from
Applied Vegetation
kong Delta, Vietnam,
culture: the structural
American Enterprise
ical guide to wavelet
ogical Society, 79, pp.
J.P., 2008. MODIS
abu, J.Y., Salas, W.,
riculture in South and
DIS images. Remote
REAL TIME SPEED ESTIMATION FROM MONOCULAR VIDEO
M. S. Temiz* *, S. Kulur"', S. Dogan*
» Ondokuz Mayis University, Dept. of Geomatic Engineering, Samsun, TURKEY — (mstemiz, sedatdo)@omu.edu.tr
^ Istanbul Technical University, Dept. of Geomatic Engineering, Istanbul, TURKEY - kulur@itu.edu.tr
Commission III, WG III/V
KEY WORDS: Video images, speed estimation, monocular video, object tracking, optical flow, traffic surveillance.
ABSTRACT:
In this paper, detailed studies have been performed for developing a real time system to be used for surveillance of the traffic flow by
using monocular video cameras to find speeds of the vehicles for secure travelling are presented. We assume that the studied road
segment is planar and straight, the camera is tilted downward a bridge and the length of one line segment in the image is known. In
order to estimate the speed of a moving vehicle from a video camera, rectification of video images is performed to eliminate the
perspective effects and then the interest region namely the ROI is determined for tracking the vehicles. Velocity vectors of a
sufficient number of reference points are identified on the image of the vehicle from each video frame. For this purpose sufficient
number of points from the vehicle is selected, and these points must be accurately tracked on at least two successive video frames. In
the second step, by using the displacement vectors of the tracked points and passed time, the velocity vectors of those points are
computed. Computed velocity vectors are defined in the video image coordinate system and displacement vectors are measured
by the means of pixel units. Then the magnitudes of the computed vectors in the image space are transformed to the object
space to find the absolute values of these magnitudes. The accuracy of the estimated speed is approximately + 1-2 km/h. In
order to solve the real time speed estimation problem, the authors have written a software system in C++ programming language.
This software system has been used for all of the computations and test applications.
1. INTRODUCTION
Using of video sequences are increasing in several applications
for automation, for instance tracking moving objects, extracting
trajectories, finding traffic intensity or estimating vehicle
velocity etc. In this paper we explain the results of our system
which we have developed for automatic real-time estimation of
the speed of moving vehicles from video camera images. This
approach requires only the knowledge of two lengths on the
ground plane, no interior or exterior calibration parameters are
required if frontal image acquisition plan is assumed. So we
assume that the rest of interesting road segment is planar and
straight and the camera is fixed on the ground. Our system can
determine vehicle speed in real time and with high accuracy by
using any kind of digital camera. This procedure involves two
main steps to be solved. At the first step, enough number of
points from the vehicle is to be selected, and these points should
be tracked at least on two successive video frames. At the
second step, by using the displacement vectors of the tracked
points and passed time, velocity vectors of those points are
computed. Due to the nature of the images’ perspective effects,
the certain geometric properties of the scene such as lengths,
angels and area ratios are distorted. These distortions must be
corrected. At first, the background image is detected by using
background extraction methods and lines on the images are
detected automatically with Hough Transformation approach. In
order to rectify images of the scene, we use vanishing point
geometry and thus solve the scale problem. Vanishing points
are automatically detected with those extracted lines by using
least squares adjustment. Subsequently a projective
transformation is applied to rectify images by using these
vanishing points. Actually there is no need to apply projective
transformation for all over the image for rectification. Instead of
rectifying the whole image, we only rectify the values of the
* Corresponding author.
distorted velocity vectors and thus gain time for real time
computations.
In order to track moving objects from video images, the points
to be tracked which belong to the moving object on the
successive video frames should be selected automatically. For
this purpose we use corner detection algorithms to
automatically select those points.
In the literature, generally two methods are used for tracking the
selected points. Maduro et al. (2008), have used Kalman
filtering method for tracking points on the subsequent frames to
estimate the velocities of the vehicles, and they have reported
2% accuracy. In the similar manner, Li-Qun et al. (1992), Jung
and Ho (1999), Melo et al. (2004) and Hu et al. (2008b) have
also used the Kalman filtering method for tracking the selected
points. The other method used for tracking the selected points is
optical flow. Sand and Teller (2004), Sinha et al. (2009), Dogan
et al. (2010) and Santoro et al. (2010), have used optical flow
method for tracking points in subsequent frames.
For tracking of the selected points we use Lukas — Kanade (LK)
Optical Flow approach. In order to test the system, we have
monitored a car which moves with a GPS receiver to measure
its speed by GPS technique and compared the GPS speed values
to the values of our video camera speed estimation system and
we obtained the vehicle speed within + 1 km/h accuracy. In this
paper, we explain the determination of the vehicle’s speed and
we give sample applications selected from our test studies. We
also explain the approaches and mathematical models that we
used for the solution of the problem. Solutions and the models
to be used for speed estimation problem vary according to the
applications and their final purposes. When applications related
to vehicle speed estimation problems are investigated, two main
fields are distinguished: traffic surveillance (Gupte, et al., 2002)
