Page 1

REAL-TIME ROAD TRAFFIC MONITORING USING A FAST A PRIORI KNOWLEDGE

BASED SAR-GMTI ALGORITHM

Stefan V. Baumgartner, Gerhard Krieger

Microwaves and Radar Institute, German Aerospace Center (DLR)

Muenchner Strasse 20, 82234 Wessling, GERMANY, Email: stefan.baumgartner@dlr.de

ABSTRACT

Radar systems operating on high altitude platforms can pro-

vide traffic information over wide areas, independent of sun-

light illumination and weather conditions. In the paper, a

novel a priori knowledge based ground moving target indi-

cation (GMTI) and parameter estimation algorithm applica-

ble on single- as well as on multi-channel synthetic aperture

radar (SAR) data is presented. Only the intersection points

of the moving vehicle signals with the a priori known road

axes, which are mapped into the range-compressed data do-

main, are evaluated. The algorithm needs low computational

load and is hence well suited for real-time traffic monitoring

applications.

Index Terms— Synthetic aperture radar, pulse Doppler

radar, radar signal processing, road vehicle location

1. INTRODUCTION

Nowadays, a lot of motorways are equipped with sensors to

monitor the actual traffic situation with the aim to ensure mo-

bility (avoid congestions) and to increase the safety of road

users. Unfortunately, suchdetailedtrafficinformationismiss-

ing outside the major motorways due to the lack of sensor

installations. Radars flying at high altitudes provide an ele-

gant solution to fill this gap, especially if this information is

required only on a non-regular basis as in the case of spe-

cial events or catastrophes. For this, a new radar based traffic

monitoring system is currently being developed by the Mi-

crowaves and Radar Institute of the German Aerospace Cen-

ter (DLR). This airborne system has the challenging task to

acquire, process and deliver the relevant traffic products to a

dedicated traffic management center in real-time. SAR and

GMTI processing have to be carried out directly onboard the

aircraft (cf. Fig. 1). Due to bandwidth limitations, only the

relevant traffic data are transmitted to a ground station us-

ing a laser communication terminal or a microwave downlink.

After further processing the data are forwarded to the traffic

management center.

Principally already existing GMTI systems and algo-

rithms originated in the military field can be used for moving

vehicle detection and parameter estimation. However, most

F-SAR

onboard SAR-GMTI

processor

data downlink

Aircraft

further processing

data distribution

Ground Station

Traffic Management

Center

Fig. 1. Radar based traffic monitoring concept.

of these algorithms require large computing power and, if the

computation should be performed in real time, the system

complexity and the costs become astronomical. For traffic

monitoring applications each vehicle has to be assigned to

a certain road. For this task anyway a road data base is re-

quired. Furthermore, it is not necessary to detect vehicles

moving off-road. Hence, by incorporating the a priori known

road network already into the detection stage of the GMTI

algorithm and by ignoring off-road moving vehicles, the sys-

tem complexity, the costs as well as the computational load

can be reduced significantly.

The idea to use a road network is not new, but up to now

the road network mainly was used together with displacement

based GMTI algorithms. These algorithms measure the az-

imuth displacements of the vehicles, occurred due to conven-

tional SAR focusing, for computing the across-track veloci-

ties [1]. The required processing is time consuming since in

general SAR images have to be generated taking into account

the full bandwidth given by the pulse repetition frequency.

Our proposed algorithm does not require SAR focus-

ing, since it operates on single- or multi-channel range-

compressed SAR data. The geocoded position of each de-

tected moving vehicle is directly obtained from the inter-

section of the road axis with the range-compressed moving

vehicle signal. Motion parameter computation is done by

estimating the Doppler frequency of the signal at the road

intersection. The parameters absolute velocity, heading and

geocoded position can be estimated with high accuracy.

Page 2

2. ALGORITHM

2.1. Principle

As a first step the a priori known road axis is mapped into the

range-compressed SAR data array. The required coordinate

transformation, which is the heart of the proposed algorithm,

is done such that the geographical coordinates of each road

point are transformed to corresponding beam center coordi-

nates in the range/azimuth plane. The beam center position

of a detected moving vehicle is then given by the intersection

of the vehicle signal with the mapped road point (cf. Fig. 2).

range

azimuth

road beam

center coordinates

moving

target signal

intersection

azimuth samples @ intersection

data array

Fig. 2. Principle of the proposed algorithm.

Owing to the mapping the geographical coordinates of the

road point and, hence, the coordinates of the detected vehicle

moving on this road point at beam center time tbcare known,

so that no further geocoding is required. For moving vehi-

cle detection and motion parameter estimation only a few az-

imuthsamplesaroundtheintersectionpointaretaken(cf. Fig.

2 right) and transformed into Doppler domain via FFT. Due to

the small number of used azimuth samples, the signal phase

is more or less linear over time and so the moving vehicle

signal appears as a sharp peak in Doppler domain. For detec-

tion the signal amplitude is compared to a certain threshold

and for motion parameter estimation the Doppler shift fDC

of the signal peak is exploited. The proposed algorithm is

well suited for airborne but not for spaceborne applications,

since the detection performance suffers from low SNR.

2.2. Structure of the Algorithm

In Fig. 3 the flow chart of the proposed GMTI algorithm

is shown exemplarily for a dual-channel system. RX1 and

RX2 are the range-compressed images. Clutter suppression is

performed using the displaced phase center antenna (DPCA)

technique. The geographical coordinates as well as the el-

evations of the roads of interest are obtained from a road

database. Interpolation of these coordinates is necessary to

avoid gaps in the range/azimuth plane.

Around each road point in the range/azimuth plane some

azimuth samples are extracted from the DPCA data and trans-

formed into Doppler domain using the FFT. Each detected

signal peak in the Doppler domain corresponds to a potential

moving vehicle. The parameter estimation procedure is ex-

plained in section 2.4. Before formatting and distributing the

RX2

Baseline Estimation,

Co-Registration &

Channel Balancing

RX1

RX 2

co-reg.

RX 1

DPCA

Extract Azimuth Samples

Around Road Point

repeat for every road point

da⇐

FFT

Detection

Motion Parameter

Estimation

Clustering

Formatting

Distribution

Navigation Data &

Radar Parameters

Roads of Interest

Interpolation

Coordinate

Transformation

Doppler

Centroid

Estimation

repeat for every road of interest

fDC,st

x0, r0,

r10, tbc,

α

vp,hp, αp,

ts, λ

fDC

geographical coordinates

in UTM (northing, easting,

elevation)

Fig. 3. Simplified flow chart of the proposed algorithm.

data to the traffic management center, a clustering operation

is performed, where multiple detections of one and the same

vehicle are merged to only one physical vehicle (cf. Fig. 3

bottom). The whole algorithm sketched in Fig. 3 can also

be used for single-channel systems by just omitting the stages

”Baseline Estimation, ...” and ”DPCA”. With single-channel

systems only fast moving vehicles falling outside the clutter

band are detectable, but for these fast vehicles the parameters

absolute velocity, heading and geocoded position can be es-

timated with high accuracy. Furthermore, instead of having

only two channels also multiple channels and more sophisti-

cated techniques like space-time adaptive processing [2] can

be applied at the cost of increased computational load.

2.3. Coordinate Transformation

The relation of the global Cartesian UTM coordinate system

{xUTM,yUTM,zUTM} and the local Cartesian coordinate

system {x,y,z} relevant for GMTI processing is sketched in

Fig. 4. The x-axis is defined by the platform velocity vector

? vp, which is assumed to be constant. A squinted geometry

has to be considered, since in general it can not be ensured

that the squint angle and, hence, the Doppler centroid of the

clutter is negligibly small. In Fig. 5 it is shown how the re-

ceived non-squinted and squinted data of one and the same

stationary road point are stored in the range-compressed SAR

data array. In Fig. 5 xr0is the azimuth position of the road

point at minimum range r0, x0is the azimuth offset due to

Page 3

xUTM(easting)

x

y

z

flight

direction

vp

yUTM(northing)

zUTM(elevation)

N

S

EW

αp

α

α

αr

90°

road

tangent

on road

road point on ground

@ beam center

αp

Fig. 4. Relation between the global geographical UTM and

the local Cartesian coordinate system.

x0

x

azimuth

range

r0

r10

TSA

TSA

xr0

vp

vp

xpt

rf

Fig. 5. Range-compressed SAR data array containing a single

road point in the non-squinted (left) and squinted (right) case.

the squint angle and r10is the beam center range. By decom-

posing the range vector ? r into a component parallel and into

a component perpendicular to flight direction, the vectors ? xr0

and ? r0are obtained:

? xr0(t = ts) = ?? vp,? r(t = ts)?

? vp

?? vp?2,

(1)

where tsis the absolute start time of data acquisition, ?.? is

the inner product and ?.? is the L2norm. The vectors ? vpand

? r can be computed using the known UTM coordinates of the

stationary road point and the radar platform. The minimum

range r0is then given by

r0= ?? r(t = ts) − ? xr0(t = ts)?.

(2)

For computing the azimuth position xpt of the road point

within the data array the following equation can be used:

xpt=

?

? vp

?? vp?,

? xr0(t = ts)

?? xr0(t = ts)?

?

?? xr0(t = ts)? − x0.

(3)

The azimuth offset x0can be computed as

x0= r0tanψ,

(4)

where the squint angle ψ is given by

ψ = arcsin

?λfDC,st

2vp

?

= arccos

?r0

r10

?

.

(5)

In previous equation λ is the radar wavelength and fDC,stthe

Doppler centroid of the clutter, which can be estimated from

the data of a single channel. Knowing the squint angle the

beam center range can be computed:

r10=

r0

cosψ.

(6)

The beam center time of the road point is given as

tbc= ts+xpt

vp

.

(7)

2.4. Motion Parameter Estimation

The motion equations of a vehicle moving at constant altitude

hvcan be written as

xv= xpt+ v0(t − tbc)cosα +1

2a(t − tbc)2cosα,

(8)

yv= y0+ v0(t − tbc)sinα +1

2a(t − tbc)2sinα,

(9)

where v0is the absolute velocity at beam center time tbc, a is

the constant acceleration and α is the road angle with respect

to the x-axis (cf. Fig. 4). The across-track position of the

target at t = tbcis denoted as y0and given as

y0=

?

r2

0− ∆h2,

(10)

where ∆h = hv− hpis the altitude difference between the

moving vehicle and the radar platform. The distance from the

transmit antenna to the target is then

r(t) =

?

[xv− xpt+ x0− vp(t − tbc)]2+ y2

v+ ∆h2.

(11)

After performing a second order Taylor expansion and some

substitutions the range can be approximated as [3]

r(t)∼= r10−λ

2fDC· (t − tbc) −λ

4ka· (t − tbc)2

(12)

where fDC is the total Doppler shift of the received signal

due to squint and target motion and kais the Doppler slope.

The Doppler shift fDC can be estimated after transforming

the azimuth samples around the road intersection point (cf.

Fig. 2 right) into Doppler domain. The absolute beam center

vehicle velocity can then be computed as

v0=

????

λr10(fDC,st− fDC)

2(x0cosα + y0sinα)

????= |vabs|.

(13)

The heading of the vehicle is given by

αv=

?

α

if sgn(vabs) = +1

if sgn(vabs) = −1

α − 180◦

,

(14)

where sgn(.) is the signum function.

Page 4

3. EXPERIMENTAL DATA

In 2007 several GMTI experiments have been conducted us-

ing DLR’s new F-SAR system [4]. As test sites an airfield in

Memmingen and a region around the Chiemsee, both located

in Germany, have been used. F-SAR has been operated in X-

band in a dual-channel mode. Some of the controlled vehicles

were equipped with GPS to gain reference positions and ve-

locities for the GMTI algorithm verification. Simultaneously

with the radar optical images from the same scene were taken.

In Fig. 6 the obtained GMTI results from a data take acquired

detail

range

azimuth

range

az.

range

az.

Fig. 6.

compressed DPCA image of the ”detail” with overlaid run-

way axis and detected moving vehicles as triangles (bottom

left) and corresponding SAR image (bottom right).

SAR image of Memmingen airfield (top), range-

over the Memmingen airfield are shown. All controlled vehi-

cles have moved on the runway in across-track direction. The

estimated velocities are 8.6, 84.2, 14.2 and 42.7 km/h (128

azimuth samples were used for parameter estimation). Com-

pared to the reference data the velocity estimation errors are

-1.5, 3.5, -1.8 and -1.3 km/h. The position errors are 17.9, 9.9,

17.3 and 16.5 m. The runway in Memmingen is about 30 m

broad and as road axis for the coordinate transformation the

middle of the runway was chosen, but during the experiment

the vehicles have moved on the edge. Under this aspect, the

estimation accuracy is quite good. In the ”Formatting” stage

also KML files are produced, which easily can be visualized

using Google Earth as shown in Fig. 7. Here a GMTI result of

a Chiemsee data take, where a lot of customary road vehicles

have been detected on the autobahn A8, is visualized.

4. CONCLUSIONS

A GMTI algorithm suitable for single- and multi-channel

SAR data based on a priori knowledge was presented. The

algorithm was verified using real dual-channel SAR data ac-

quired with DLR’s airborne system F-SAR. The obtained

performance implies that the algorithm is applicable for real-

time traffic monitoring applications.

Fig. 7. Quick-look F-SAR image (size 1.5 x 0.9 km) of the

autobahn A8 as Google Earth overlay. The shown vehicles

were automatically detected and their parameters were auto-

matically estimated using the proposed algorithm.

5. REFERENCES

[1] F. Meyer, S. Hinz, A. Laika, D. Weihing, and R. Bamler,

“Performance analysis of the TerraSAR-X traffic moni-

toring concept,” ISPRS Journal of Photogrammetry and

Remote Sensing, vol. 61, no. 3-4, pp. 225–242, 2006.

[2] R. Klemm, Space-Time Adaptive Processing: Principles

and Applications. UK: IEE, 1998.

[3] S. V. Baumgartner and G. Krieger, “SAR Traffic Moni-

toring Using Time-Frequency Analysis for Detection and

ParameterEstimation,” inIEEEInternationalGeoscience

and Remote Sensing Symposium (IGARSS), vol. 2,

Boston, USA, July 2008, pp. II–25 – II–28.

[4] R. Horn, A. Nottensteiner, and R. Scheiber, “F-SAR -

DLR’s advanced airborne SAR system onboard DO228,”

in7thEuropeanConferenceonSyntheticApertureRadar,

vol. 4, Friedrichshafen, Germany, June 2008, pp. 195–

198.