In: Wagner W., Székely, B. (eds.): ISPRS ТС VII Symposium - 100 Years ISPRS, Vienna, Austria, July 5-7, 2010, IAPRS, Vol. XXXVIII, Part 7B
470
The present publication shows a vehicle detection approach based
on the Viola-Jones detector (Viola and Jones, 2004) trained by
Gentle AdaBoost (Friedman et al., 2000). Vehicle detectors based
on Boosting had already been applied to aerial images success
fully in literature (e.g. Nguyen et al., 2007; Grabner et al., 2008).
In the optical system for online traffic monitoring the DLR in-
house developed sensor called 3 K camera (e.g. Kurz et al., 2007)
is included, which is capable of direct orthorectification / georef-
erencing in conjunction with a IGI lid Aerocontrol RT IMU/GPS
navigation system and several PCs for image processing. Fur
thermore, the sensor provides a high image repetition rate of up
to 3 Hz which offers a high overlap of sequential images and
makes vehicle tracking possible. With a big footprint of 4 km
across track at a typical flight level of 1500 m over ground and
a high native resolution of 20 cm GSD (Nadir) it is well suited
for recording road traffic data. After several years of develop
ment our system has reached an operational state and is ready for
application in disasters and mass events.
The paper is structured as follows. Second section presents the
system for airborne traffic monitoring and the data sets obtained.
Then, the processing chain for automatic road traffic data extrac
tion from aerial images is introduced in section 3. Section 4 deals
with the validation of the processing chain, and the last section
gives conclusions and presents the plans for future work.
2 SYSTEM AND DATA SETS
The real-time road traffic monitoring system consists of two parts.
One part is installed onboard the aircraft, consisting of the 3 K
camera system, a real-time GPS/IMU unit, one PC for each sin
gle camera processing image data, one PC for traffic monitoring
tasks, a downlink-antenna with a band width of 5 Mbit/s (actual,
upgradeable to a bandwidth of about 20 Mbit/s) automatically
tracking the ground station, and a PC for steering the antenna
and feeding the downlink with data. The ground station mainly
consists of a parabolic receiving antenna, which is automatically
aligned with the antenna at the aircraft, and a PC system for vi
sualization of the downlinked images and traffic data. Given an
internet access at the place of the ground station, the obtained
traffic data will be directly transferred to an internet traffic portal.
2.1 Onboard System
The system for traffic monitoring aboard the airplane uses aerial
image sequences obtained with the so called 3 K camera system.
All results shown in this publication are based on this sensor. In
near future this optical sensor will be replaced by the successor,
the 3 K+ camera system. Both wide area digital frame camera
systems offer similar properties (Tab. 1). Each consists of three
non-metric Canon EOS IDs cameras mounted on a ZEISS aerial
platform. One look in nadir direction and two looks in oblique
sideward direction result in an increased FOV of up to 104 degree
or 31 degree in side resp. flight direction based on a focal length
of 50 mm. The cameras acquire images with a frame rate of up
to 5Hz; here the absolute number of images is limited due to an
overflow of the internal memory. Based on the image size of up to
21 MPix and a colour depth of 24 bits the overall output data rate
of the three camera system lay between 6 and 10 MByte/s for jpeg
compressed images. The data rates at the cameras depend also on
the flight and image acquisition mode, e.g. the overlap and the
flight height. The image acquisition geometry of the DLR 3 K
and 3 K+ camera system are similar except the smaller GSD of
the 3 K+ system. Typical flight heights of the camera systems are
between 500 m and 3000 m above ground.
3K camera system
3K1 camera system
Cameras
3 x FOS IDs Mark I!
3 X EOS IDs Mark III
Image size
4992 x3328 ( lô.TMPix)
5616 x 3744 (21.0 MPixl
Max. frame rate
3 I-Iz (- 50 images )
5IIz (63images )
File size
20MByte (RAW)
25MBvte (RAW)
5,5MByte (JPEG level 8)
6,5 MByte (JPEG level 8)
ISO
100- 1600
50 - .3200
Aperture
1.4 22
1.4-22
Lenses
Canon EF 1.4 50mm
Zeiss Makro-Planar 2/50mm
Tilt angle of sideward cameras
Max 32° /variable
Max 32° / variable
Data rates (3 cameras. JPEG):
60% overlap, 1000m a.g.. 2s
3xbursts (2Hz). 1000m a.g., 6s
pause (traffic modus)
8.3 MByte's
6,6 MByte/s
9.8MByte/s
7,8MByte/s
Interface
Firewire IEEE 1394a
USB 2.0
Image sensor
full frame CMOS sensor
full frame CMOS sensor
Pixel size
7.21pm
6.41pm
Footprint /GSD, 1000m a.g.
2560m x 4S0m / 15cm nadir
2560m x 4$0m 13cm nadir
Footprint GSD, 3000m a.g.
7680m x 1440m 45cm nadir
7680m x 1440m / 39cm nadir
FOV (side resp. in fliuht)
104*731°
104=731°
Calibration (interior orientation)
5 parameters:
focal length, focal point, radial
distortion Ai and A2
Not calibrated yet.
GPS/IMU accuracies
IGI lid
IGI lid
postproc. (ip/yaw/xy/z)
0.003*70.007*70.08m/0.05m
0.003 < 70.007 < 70.08m.'0.05m
real lime (rp/yaw/xy/z)
0.01 °/0.05=70.1 m/0,1 m
0.01 *70.05=70.1 m/0.1 m
Georeferencing accuracies
Direct georeferencing
plus bundle adjustment
2-4m
<lm
Not tested yet.
Depending on the motif and other configuration parameters
Table 1; Overview about 3K and 3K+ camera system.
The high input data rate on the one hand and the processing inten
sive modules on the other hand put high demands on the on-board
image processing hardware, which consequently leads to a multi
host solution with five PCs in total (Fig. 1). All of them run
32bit-Windows XP due to the fact that some of the third-party
software we use in our processing system only supports Win
dows. Each camera is connected via Firewire IEEE 1394a to a
dedicated host. It streams the images directly without memory-
card buffering to the camera PCs (PCI - PC3). The EOS Digital
Camera Software Development Kit (EDSDK) is installed on each
of these hosts and provides a C language interface for the control
of the cameras and the download of images to the host PC. Sup
ported operating systems are Microsoft Windows 2000 or higher
and Mac OS X since version 10.4. Since the orthorectification
and georeferencing process needs the exact position and orienta
tion of the airplane, the IGI lid GPS/IMU system is connected via
Ethernet to the onboard system. The fibre-optic gyro based Iner
tial Measurement Unit with its integrated 12-channel L1/L2 GPS
receiver are triggered by the cameras external flash signal. Ev
ery time a flash signal is received, the IGI lid sends coordinates
and orientation via a TCP connection to one of the camera PCs.
A module runs a TCP client and matches the received geo-data
with the image received from the camera. The image is written to
disk where it can be read by the orthorectification module. The
geo data is sent to this module via message passing. After the or
thorectification process (Muller et al., 2002) has completed, the
orthorectified image is written to disk, copied to PC 4 and could
be processed for automatic traffic data extraction or sent down to
ground station.
The onboard system was ready to fly in mid 2009. Unfortunately
the assignation of the permit to fly certificate of this system on
board the DLR aircraft Cessna 208 B was delayed. We got the
certificate in April 2010 and we are currently carrying out series
of test flights. During first flights, several problems have been
exposed and fixed as well as an optimal start-up procedure for
hardware and software of this complex system had to be found.
Meanwhile the sensor and system components for orthorectifi
cation and automatic traffic data extraction are running properly
during test flights and we recorded road traffic data aboard the
aircraft during flights successfully. Due to a hardware defect of
the S-band downlink, we have not yet transmitted traffic data to
the ground, but we are sure to catch up on this very soon. Al
though the onboard traffic data processing is already working, re-
