09-05-2012, 10:58 AM
Development of Detection Algorithm for Vehicles Using Multi-line CCD Sensor
Development of Detection Algorithm for Vehicles Using Multi-line CCD Sensor.pdf (Size: 267.46 KB / Downloads: 25)
Abstract
We have developed a lane change aid system (LCASj,
which detects vehicles behind in adjacent lanes with
multi-line CCD sensors and informs the driver of vehicle
location with a head-up display (HUDj. Instead of
processing camera images direct&, a multi-line CCD
sensor contains the pairs of line CCDs and measures Iwodimensional
distance distribution by comparing the
brightness on line CCDs. To ensure the efffctive vehicle
detection in real public trafic, we have developed a
detection algorithm that evaluates the reliabilify of each
measured distance and distinguishes vehicles from
guardrails and white markers on roads. We will show
the detection algorithm and its performance in real trafic.
1. Introduction
It has become more crucial to detect the driving
environment as the researches and developments on active
safety systems for automobiles have been frequently made.
As for the sensing devices for driving environment, the
applications of image processing using cameras have also
become known in addition to laser radar, milliwave radar,
and ultrasonic sensors [l] [2] [3].
Based on the fact that drivers watch side mirrors
repeatedly before they change lanes, we have developed
an LCAS, which reduces the stress of drivers at lane
change by informing them of the location of vehicles
Multi-line CCD sensor
HUD
Figure 1. System configuration
behind in the adjacent lanes [4]. Our LCAS with multiline
CCD sensors is illustrated in Figure I.
The sensors in side mirrors observe the adjacent lanes
through half mirrors and measure the distances to objects
by comparing the brightness distributions on the pairs of
line CCDs inside. Because LCAS is bascd on a simpler
measurement principle than the ordinary image
processing using cameras, we can reduce the size and
complexity of both the sensor itself and processing unit.
The sensor, however, detects every object with distinct
brightness such as guardrails and markers on roads that
never disturb lane change maneuvering. It is necessary
to establish a detection algorithm which enables LCAS to
detect only vehicles in adjacent lanes.
We will describe the outline of the multi-line CCD
sensor, our detection algorithm, and the results of
application of the multi-line CCD sensor to real traffic.
2. Multi-line CCD sensor
We will summarize the outline of a multi-lins CCD
sensor and its principle of distance measurement. The
results of vehicle measurement will be shown as well.
2.1. Outline of multi-line CCD sensor
The multi-line CCD sensor is small enough to be
installed in the side mirrors of a vehicle as shown in
Figure 2. There are 16 pairs of line CCDs in the sensor.
One line CCD of each pair is located behind the upper
Figure 2. Multi-line CCD sensor
21
0-7803-5467-2/99$/1 0.00 0 1999 IEEE
lens, and the other behind the lower one. The parallax
between the brightness distribution on each pair of line
CCDs can be converted into distance. Figure 3 shows
the brightness distribution of the line CCDs when a
flashlight is put on the ground in front of the sensor at
night. In Figures 3 (a) and (b), a few upper and lower
line CCDs show significant brightness due to the
flashlight. The brightness distributions on the eighth
line CCDs are shown in Figure 3 ©. The parallax
represents the distance from the sensor to the flashlight.
2.2. Applications to measurement of vehicles
In the applications of the multi-line CCD sensor to
vehicle detection, each line CCD is virtually divided into
46 even windows, and a distance is calculated on the
brightness distribution on each window. Figure 4 shows
an example of detecting a vehicle 20 m away. Figure 4
(a) is the video recorded image, and (b) the measured
distance of each line and window. The sensor yields
255 I n Ill1
vv))
E
E
rn
U
'C
n-
Pixel number on line CCD 511
Figure 3 (a). Brightness on upper CCDs
Line number n (v0)
ac,
0
'C
E
m
0
Pixel number on line CCD
Figure 3 (b). Brightness on lower CCDs
Pixel number on line CCD 511
Figure 3 ©. Brightness on 8th line CCDs
precise two-dimensional distance distribution for a
vehicle. The result of vehicle measuement at various
distances is shown in Figure 5. Though there is an error
of about 10 % at 50 m, it can be compensated by
adjusting the sensor gain in advance because the deviation
of the measurement is rather small.
Figure 4 (a). Recorded image of a vehicle
Line number
Figure 4 (b). Measured distances of a vehicle
I I
0 10 20 30 40 50
Real distance (m)
Figure 5. Distance measurement of vehicle
22
3. Detection algorithm
Based on the parallax, we can get the distances to
objects easily. It is, however, still necessav to develop
an effective detection algorithm for vehicles in real traffic.
we will describe a problem of multi-line CCD sensor in
real traffic and show the detection algorithm.
3.1. Problem in applying to real traffic
Figure 6 shows the result of the detection of a vehicle
in the right adjacent lane in real traffic. Figure 6 (a) is
the recorded image, and (b) the detected distances. The
markers on the road are also detected. In applying this
sensor to real traffic, the unnecessary information such as
markers on roads should be deleted.
3.2. Algorithm design
To remove the information of markers on the road, we
evaluate the three dimensional arrangement of detected
Figure 6 (a). Recorded image of real traffic
Line number
Figure 6 (b). Measured distances of real traffic
data. Figure 7 illustrates the horizontal and vertical
distributions of measured data with the sensor. We can
calculate the horizontal distribution from the measured
distances and their corresponding line locations, and the
vertical distribution from the distances and their window
locations. Each notation in the figures represents each
line number. As seen in Figure 7 (b), the markers on the
road as well as the vehicle are above the horizontal plain
because the vehicle behind is on a downhill. From
Figures 7 (a) and @), a vehicle can be determined as an
object that is located within about 3.5 in laterally and
contains vertically aligned distance data.
Based on the above characteristics of vehicle, we first
eliminate the markers on roads, or the distance data which
align on a horizontal plain, and then extract the
information of vehicle. To simply implement the
concept of eliminating markers on roads, we apply the
following process to the data evaluation.
Let (x,,z,) and (x,+,,z8+,) he the measured distances in
a certain line CCD, while n and z ai% the longitudinal and
vertical positions, respectively, and the suffix, i, denotes
the window number. Both of data will he eliminated
when they satisfy the following condition.