Progress In Electromagnetics Research, Vol.

148, 114, 2014

Analysis of Moving Human Micro-Doppler Signature

in Forest Environments
Jose M. Garcia-Rubia* , Ozlem Kilic, Vinh Dang, Quang Nguyen, and Nghia Tran

AbstractAutomatic detection of human motion is important for security and surveillance

applications. Compared to other sensors, radar sensors present advantages for human motion detection
and identification because of their all-weather and day-and-night capabilities, as well as the fact that
they detect targets at a long range. This is particularly advantageous in the case of remote and highly
cluttered radar scenes. The objective of this paper is to investigate human motion in highly cluttered
forest medium to observe the characteristics of the received Doppler signature from the scene. For this
purpose we attempt to develop an accurate model accounting for the key contributions to the Doppler
signature for the human motion in a forest environment. Analytical techniques are combined with full
wave numerical methods such as Method of Moments (MoM) enhanced with Fast Multipole Method
(FMM) to achieve a realistic representation of the signature from the scene. Mutual interactions between
the forest and the human as well as the attenuation due to the vegetation are accounted for. Due to
the large problem size, parallel programming techniques that utilize a Graphics Processing Unit (GPU)
based cluster are used.

1. INTRODUCTION
A capability to detect, track and monitor human motion in highly cluttered environments, such
as forests, is important for security and surveillance operations. Particularly for border control
applications, it is essential to track and classify humans moving in remote locations over a long period
of time, at a low cost, and with low maintenance. Several different radio-frequency devices have been
developed for this purpose; in particular Doppler radars are widely used for human discrimination
applications [1, 2].
Human movement can be classified by using micro-Doppler radar systems that are capable of
detecting small variations in the carrier frequency caused by motion. Studies have shown that the human
micro-Doppler signature is a unique signature, differentiable from all other micro-Doppler signatures,
including those caused by four-legged animals [3, 4]. The micro-Doppler frequency shift depends on
the transmitted frequency and the velocity of the different body parts over time [58]. This has
been exploited for identification and classification of different human activities [9]. To analyze the
micro-Doppler signatures, a joint time-frequency transform such as the Short-Time Fourier Transform
(STFT) is performed. Different types of motions can be identified and classified from micro-Doppler
spectrograms [9, 10].
Most of the recent research in this field have focused on using continuous wave (CW) radar for
Doppler detection due to their simpler system design and implementation, and lower development cost.
For example, a CW microwave Doppler radar operating at 2.4 GHz was developed for multiple-mover
sensing [11]. CW microwave Doppler radars have been widely used for vital sign detection and life
Received 23 January 2014, Accepted 21 May 2014, Scheduled 15 June 2014
* Corresponding author: Jose Miguel Garcia-Rubia ([email protected]).
The authors are with the Department of Electrical Engineering and Computer Science, The Catholic University of America, 620
Michigan Ave., N.E., DC 20064, USA.

Garcia-Rubia et al.

detection [1215], and for extracting the Doppler signatures for biometric characterization [16]. Signal
processing and Doppler extraction method for CW Doppler radar were also discussed in [4, 1722].
The objective of this paper is to investigate human motion in a highly cluttered forest medium to
observe the characteristics of the received Doppler signature from the scene. We present the results
of an accurate model which accounts for the key contributions to the Doppler signature of the human
motion in a forest environment.
A CW radar is assumed to generate a plane wave that is incident on the scene. The incident
wave is attenuated as it propagates through the forest based on reported specific attenuation values
for wooded areas [23]. Human motion is created within the scene to analyze the received signature
characteristics. The backscattered fields at the radar are calculated from the contributions of the forest
and the human in motion accounting for the first-order mutual coupling effects between the forest and
the human. The forest scattering model represents the trees as PEC cylinders distributed randomly
over a flat ground, and employs a first-order approximation where the contributions from each tree
are summed independently. A full wave technique, namely Method of Moments enhanced with Fast
Multipole Method (MoM-FMM), is employed to accurately model the scattered fields from the human
in motion since this is the main term which contributes to the Doppler signature. The mutual coupling
between human and the forest can also contribute to the information about the human in motion, and
is included in the model.
The rest of the paper is organized as follows. In Section 2, the human walking model based on
empirical data is presented. The scattering models for the human and forest are explained in Section 3.
The micro-Doppler radar spectrograms are presented in Section 4. The simulation results for various
test case scenarios are presented in Section 5. Finally, the conclusions are summarized in Section 6.
2. HUMAN WALKING MODEL
Realistic human motion models have been investigated by many authors. Some of these models are
obtained from real motion capture data [24]. However such data is limited to specific parameters from
the captured scene such as the particular direction, speed, and human characteristics. In order to study
different human motion effects, we opt to use a standard analytical model extracted from empirical
data, such as the well-known Boulic model [2528].
According to the Boulic model, the human body can be described by 16 joints, which are identified
by the small circles as shown in Figure 1. The body parts are defined by two joints at each end; e.g., foot
is defined by the toe and the ankle. The 17 body parts are depicted by the head and the lines between
the joints as depicted in Figure 1. Each body part is modeled using ellipsoids with different semi-axes
lengths appropriate for the corresponding body part. The motion is described using the Boulic walking
model based on time-dependent translations and rotations of each joint [25]. The input variables for
this model are the velocity and the height of the human. Given these two inputs, the human is created,
and motion characteristics are defined, as shown in Figure 1.
The motion is characterized by cycles that repeat for a constant velocity. Each cycle carries features
due to the relative motion of the various joints and body parts. The next position and orientation of a
body part in the cycle is calculated using 3 translations and 14 rotation trajectories, five of which are
duplicated for both sides of the body. The details of the motion parameters are listed in Table 1 where
Spatial and Temporal
Characteristics Lc , T c
Input: Velocity v, Height of
human H

Translations and rotations of joints

Update a position of every joint at
every time t

16 joints

Figure 1. Human motion model with 17 body parts, 3 translation trajectories and 14 rotation
trajectories.

Progress In Electromagnetics Research, Vol. 148, 2014

Table 1. Body trajectories.

Translation
(m)
General
Vertical Translation
Lateral Translation
Forward/Backward Translation
Lower body
Rotation at Ankle
Rotation at Knee
Rotation at Hip
Rotation Left/Right
Rotation Torsion
Upper body
Rotation at Elbow
Rotation at Shoulder
Rotation Forward/Backward
Rotation Thorax

Body
rotation
(deg)

Translation
Lateral

A (t)
K (t)
H (t)

A (t + 0.5)
K (t + 0.5)
H (t + 0.5)

E (t)
S (t)

E (t + 0.5)
S (t + 0.5)

LR (t)
T O (t)

F B (t)
T H (t)
Upper Body
Motion
Translation
Forward/Backward

Rotation
Ankle
Rotation
Knee
Rotation
Hip
Rotation
Left/Right

Right
rotation
(deg)

TV (t)
TL (t)
TF B (t)

Lower Body
Motion
Translation
Vertical

Left
rotation
(deg)

Rotation
Torsion

Translation
Vertical

Translation
Lateral

Translation
Forward/Backward

Rotation
Elbow

Duration of
cycle
Tc

Rotation
Shoulder

Rotation
Forward/Backward

Rotation
Thorax

Figure 2. Lower and upper body motion.

t is the relative time with respect to the duration of the cycle, Tc . All the translations and rotations are
dependent on the fundamental spatial and temporal characteristics of the motion such as the relative
length, Lc , and the duration of the walking cycle, Tc .
The fundamental spatial characteristic of the human walk in the Boulic model is the relative
length
of the walking cycle, Lc , which depends on the relative velocity, Rv , such that Lc = 1.346 Rv . The
relative velocity is defined as the average walking velocity normalized by the height of the thigh, which
is approximately 50% of the height of the human. The fundamental temporal characteristic of the
Boulic model is the duration of cycle, Tc , which is a ratio of the relative length of walking cycle to the
relative velocity; i.e., Tc = Lc /Rv . These spatial and temporal characteristics are utilized to calculate
the various translations and rotations of each body part as listed in Table 1 [25].
The translations and rotations are applied for the lower and the upper parts of the body as depicted

Garcia-Rubia et al.
5

torso
left hand
right hand
left knee
right knee
left ankle
right ankle
left toe
right toe

Speed (m/s)

4
3
2
1
0
-1

0.2

0.4

0.6
0.8
Time (s)

1.2

Figure 3. Radial velocities for different body joints in one cycle.

in Figure 2. For example, to define the trajectory of one part of the lower body, 3 translations (vertical,
lateral, and forward/backward) are applied to all joints. Then three rotation motions are applied in
sequence, starting with the ankle, continuing with the knee and finishing with the hip. Finally the
two rotations left/right and torsion are applied to all lower body joints. The joints trajectories for the
upper part of the body are calculated in an analogous way.
According to this model, a human with a height of 1.8 m, moving at 0.9 m/s has a walking cycle
period, Tc = 1.346 s. The radial velocities for different body joints are shown for a single walking cycle
period in Figure 3. The lower part of the legs presents the maximum velocity over the cycle, while the
torso depicts a velocity with very low oscillations around the velocity of the translational movement of
the human.
3. RADAR SCATTERING MODEL
A continuous wave (CW) microwave radar is assumed to generate a plane wave, E0 , that is incident on
the scene. The forest is modeled with PEC cylinders representing the trunks over a flat PEC ground, as
shown in Figure 4. The incident wave attenuates as it propagates through the forest. The attenuation
inside the medium is assumed constant for all directions based on the reported specific attenuation values
for wooded areas at frequencies above 1 GHz [23]. Three contributions from the scene, as depicted in
Figure 4 by Ef -r , Eh-r and Ef -h-r , are considered for the calculation of the total backscattered fields
from the scene.
The first term, Ef -r , is the direct contribution from the forest to the radar, which involves a firstorder analytical approach accounting for the reflections from the ground. The second term, Eh-r , is
the contribution from the human, which employs the MoM-FMM full wave method. Ground reflections
and mutual coupling effects between the different body parts are accounted for in this term. The
third contribution, Ef -h-r , accounts for the first-order interactions between the forest and human by
calculating the incident fields on the human as a result of radiated fields from all trees in the forest. The

FOREST

RADAR

E0

E radar

E fhr
E fr

E fh

E hr
HUMAN

Figure 4. Contributions to radar signature from the scene.

Progress In Electromagnetics Research, Vol. 148, 2014

other mutual coupling term, i.e., the illumination of the forest due to radiated fields from the human is
considered negligible compared to Ef -h-r . As in the calculation of Ef -r and Eh-r , the reflection from the
ground and attenuation effects are included in Ef -h-r . The fields at the radar, Eradar , are calculated as
the sum of these three terms, as in (1).
Eradar = Ef -r + Eh-r + Ef -h-r

(1)

A description of the modeling approach for these three terms is provided below.
3.1. Forest to Radar Scattering, Ef -r
Electromagnetic scattering from the forest is an important remote sensing problem for agricultural and
environmental purposes [29, 30]. For the microwave radar frequencies considered in this paper, the
trunks in a typical forest are separated by multiple wavelengths. Consequently, the received fields at
the radar can be estimated by using a first-order approximation.
We employ an analytical approach, where the trees are modeled as finite length, PEC cylinders
residing vertically over a flat PEC ground as in [31, 32]. The cylinders can be of various sizes, and
are randomly distributed over the ground. We consider a plane wave illumination on the forest while
accounting for attenuation as the wave propagates inside the medium, as shown in Figure 5. The
attenuation inside the medium is assumed constant for all directions based on the reported specific
attenuation values for wooded areas at microwave frequencies [23].
Based on this approach, the total scattered field from the forest is calculated as in (2).

N
Lj jksz zj jkiz zj Djdd
ejkr X Lj j kd j

Ef -r (
r) =
e
fdd ks , ki e
e
e
sinc (kiz + ksz )
r
2
2
j=1

Lj jksz zj +jkiz zj 2jksz (zj +d) Djdr

+ sinc (kiz ksz )
fdr ks , ki e
e
e
2

Lj jkiz zj +jksz zj 2jkiz d Djrd

frd ks , ki e
e
+ sinc (kiz ksz )
2

Lj jksz zj jkiz zj 2j(kiz +ksz )d Djrr

+ sinc (kiz + ksz )
,
(2)
frr ks , ki e
e
e
e
2
where N is the number of cylinders, Lj the height for each cylinder, k the wavenumber, and the forest
specific attenuation coefficient. ki and ks are the incidence and scattering directions, respectively. fdd ,
fdr , frd , frr are field vectors as defined in [32], and Djdd , Djdr , Djrd , Djrr are the path lengths inside the
forest corresponding to these field vectors, respectively as shown in Figure 6.

RADAR E 0

E fr
E0
e
e

E 0f

(Dd)

Figure 5. Forest contribution to radar signature.

Garcia-Rubia et al.

k i

ks

k i

k s

dr

dd

Dj

Dj

zj

fdd

f dr

k i

k i ks

f rd

k s
rr

rd

Dj

zj

zj

Dj

zj

f rr

Figure 6. Four scattering contributions from the trunks in the presence of ground.
The first term shown in Figure 6, fdd , is the direct scattering term which involves the backscattered
fields from the trunk directly towards the radar. The second and third terms, fdr and frd , include a
single reflection from the ground. The last term, frr , includes two ground reflections before the scattered
fields reach the radar.
3.2. Human to Radar Scattering, Eh-r
The scattering phenomenon in human motion studies has been conventionally modeled using the
simplistic concept of point scatters. In [28, 33], a simple primitive-based prediction technique to model
human gait was proposed. These, however, do not incorporate the mutual coupling effects between
different human body parts. Another technique that has been employed is an iterative physical optics
approach [34], which accounts for mutual coupling effects and can be useful only at very high frequencies,
such as W-band. A full wave technique, such as the one employed in this paper, provides better accuracy
for more complex structures but requires a large number of computational resources. In [35], a realistic
human mesh-based model is proposed with a radar signature modeling based on Finite Difference
Time Domain (FDTD) technique. However, FDTD requires a fine mesh in the order of /50 for
accurate modeling of curved objects, which becomes prohibitive for electrically large objects such as the
human [36].
In this paper, a full wave technique, namely Method of Moments enhanced with Fast Multipole
Method (MoM-FMM) is used to model the walking human. This enables the accurate computation of
mutual coupling effects between the different moving body parts. Ground effects are also accounted for.
Furthermore, the attenuation of the incident and scattered fields as they propagate inside the medium
are also taken into consideration.
MoM-FMM is an efficient numerical method which relies on grouping sources over the scatterer
according to their proximity to each other. It utilizes the concept of near and far interactions, which
significantly reduces the complexity of MoM from O(N 3 ) to O(N 3/2 ), where N is the number of
unknowns corresponding to the number of edges of the meshed object [37]. Implementing a parallelized
version of this method on a high performance computing (HPC) platform provides a good speedup
factor for the computation time for large-scale electromagnetic scattering problems. A brief discussion
on the parallel implementation of MoM-FMM on a GPU cluster is provided below. Detailed discussions
on the implementation can be found in [38, 39].

Progress In Electromagnetics Research, Vol. 148, 2014

3.2.1. Parallel Implementation

The GPU cluster platform used in the implementation of this model consists of 13 computing nodes.
Each node has a dual 6-core 2.66 GHz Intel Xeon processor X5650, 48 GB memory along with one Nvidia
Tesla M2090 GPU running at 1.3 GHz with 6 GB of GPU memory. The nodes are interconnected through
the Infiniband interconnection. The cluster populates CUDA v4.2, and MVAPICH2 v1.8.1.
The MoM-FMM algorithm consists of three main steps: pre-processing, processing, and postprocessing, as shown in Figure 7. The main purpose of the pre-processing step is to read in the mesh
geometry, to set up the data structure, and to divide the edges into groups. The processing step includes
two phases; namely setup and linear system solution. The matrix components for near interactions,
the radiation/receive functions and the translation matrix are calculated and stored in the setup phase.
Iterative methods are employed for solving the linear system, and each iteration involves matrix-vector
multiplications (MVM), which dominate most of the computation time. The MVM for the Zfar matrix
comprises aggregation, translation, and disaggregation steps as depicted in Figure 7, while the MVM for
the Znear matrix is only an ordinary sparse matrix-vector multiplication. Finally, the post-processing
step involves calculations of electromagnetics quantities of interest.
The entire processing step is implemented on GPU using the CUDA environment. To give an
insight into the problem size, the scattered field calculations from a human of height 1.8 m at 5 GHz
using MoM-FMM, would require 99,422 triangles for a mesh size of /8. This corresponds to 149,133
unknowns to be solved for using MoM-FMM. A detailed discussion on the parallel implementation of
MoM-FMM on GPU clusters for larger problem sizes can be found in [39].
3.3. Forest-to-Human-to-Radar Scattering, Ef -h-r
The third contribution in the calculation of the total backscattered field at the radar, Ef -h-r , accounts
for the first order interactions between the forest and human. The indirect incident fields on the human
are calculated from the radiated fields from all trees in the forest towards the human. Similar to the
calculation of the Eh-r term, MoM-FMM is utilized to compute the scattered fields from the human
Pre-processing
(CPU)

Processing
(GPU)

Setup
Calculate
near
matrix
Z near

Post-processing
(CPU)

Solve linear equation

Calculate
radiation/receive
functions

Calculate
translation
matrix

Aggregation

Z far

Translation Disaggregation
Far MVM

Figure 7. MoM-FMM processing step flowchart.

i
E0
i

E fhr

E fh

i
E fh

e
di

Figure 8. Forest-to-human-to-radar scattering.

Garcia-Rubia et al.

based on these incident fields received from the forest. As before, ground reflections and attenuation
effects are accounted for as the fields propagate inside the medium. The concept for the calculation of
the Ef -h-r term is depicted in Figure 8.
The scattered fields from each tree towards the human, Efi -h , is calculated by meshing each trunk
into triangles and employing MoM-FMM. The induced currents on the trunk due to the incident field
from the radar are computed by treating the trunk as the only scatterer in the medium. Ground
reflections and attenuation effects are included as before. The induced currents on each trunk are then
reradiated towards the human. The field incident on the human from each trunk, i.e., Efi -h , is calculated
using the equivalent dipole model as reported in [4042]. The total fields on the human are calculated
by summing Efi h for all trunks, and MoM-FMM is used to calculate Ef -h-r , i.e., the scattered field
from the human due to the forest coupling.
In order to understand the mutual coupling effects between a trunk and the human, we run a few
test cases as depicted in Figure 9. The three scenarios in Figure 9 correspond to the different positions of
the human relative to a trunk located at the origin. In each case, the separation between the trunk and
human is kept constant at 3, and the trunk is illuminated by a vertically polarized plane wave incident
from the direction of the radar. The induced fields on the human due to the indirect radiation from the
trunk are shown with a color map in each plot where the color red indicates the highest intensity. It
should be noted that the fields on the human do not include the direct illumination by the plane wave,
i.e., only the coupled fields from the trunk are shown in this figure. The relative positions of the radar,
trunk and human are depicted on the right hand side for each case. Figure 9(a) corresponds to the case
where the radar, trunk and human lie on the same axis, and the human is in the forward scattering

E fh

E fh

E0
z
x
y
y

(a)
E fh

z
E0
E fh

z
x

(b)
y

E fh
E0

z
E fh
x
y

(c)

Figure 9. Induced fields on the human due to trunk for three different human positions.

Progress In Electromagnetics Research, Vol. 148, 2014

direction of the trunk. The human is along the normal direction with respect to the radar-trunk axis
in Figure 9(b). Finally, the human is positioned between the radar and trunk, along the radar-trunk
axis, corresponding to the backscattering direction of the trunk in Figure 9(c). We observe a significant
amount of field intensity on the human when it is positioned in the forward scattering direction of the
trunk, i.e., Figure 9(a). The central part of the body is illuminated by stronger fields than the rest of
the body for this case. In the other two cases, the illumination on the human is fairly low and more
uniform.
To investigate further, we show the total observed field strengths with and without mutual coupling
effects as a function of observation angle, , for the same three test cases. The fields include the total
scattering from the human to radar for this investigation. Therefore, the no-coupling case corresponds
to Eradar = Eh-r , and the coupling case represents Eradar = Eh-r + Ef -h-r . We observe that the strongest
effects happen when the human is in the forward scattering direction of the trunk, i.e., Figure 10(a).
The highest change in amplitude for this scenario is along the forward direction ( = 180 ). The
weakest coupling contribution is observed when the human is in the backscatter direction of the trunk,
i.e., Figure 10(c). There is some contribution when the human is at an orthogonal direction to the
radar-trunk axis, i.e., Figure 10(b).
0.2

No coupling
Coupling
Magnitude

0.15

E radar

(a)

0
0

30

60

90
120
phi (deg)

150

180

0.2

No coupling
Coupling

E0

0.15
Magnitude

fh

E fhr

0.1

0.05

Eradar

E0

fh

0.1

0.05

E fhr

(b)
0

30

60

90
120
phi (deg)

150

180

0.2
No coupling
Coupling

E0

Magnitude

0.15

E radar

E fhr

0.1

0.05

(c)
0

30

60

90
120
phi (deg)

150

E fh

180

Figure 10. Comparison of mutual coupling effects for the trunk-human pair for three different human
positions.

4. MICRO-DOPPLER SIGNATURE
The micro-Doppler signatures from the scene are calculated by applying a time-frequency transform,
such as the Short-Time Fourier Transform (STFT) with a Gaussian window, to the scattered fields. To
compute this transform, we need to choose a proper sampling rate, which depends on the maximum

10

Garcia-Rubia et al.
-40

-45

150

20

-50

100

10

-55

50

-60

-10

-65

-20

-70

0.5

0
-50

-60
Left foot

Right foot
-70

Left tibia

Right tibia

-150

-80

-50

-100

-75

-30
0

Doppler (Hz)

Doppler (Hz)

-40
30

0.5
Time (s)

Time (s)

(a)

-80

(b)

Figure 11. Spectrogram of walking human at 1 and 5 GHz.

Doppler shift. Since the Doppler shift depends on the rate of motion,vR , i.e., fDoppler = 2vR /c ,
the sampling rate is directly proportional to the maximum speed observed. For a maximum speed
of 5 m/s, the Doppler shift can vary significantly based on the operating frequency of the radar. For
instance, using a 5-GHz radar results in a Doppler shift of 166.66 Hz, which requires a sampling rate of
333.33 samples/s, while a 1-GHz radar would result in a Doppler shift of 33.33 Hz and 66.66 samples/s.
An example of a spectrogram for a human of height 1.8 m, walking at 0.9 m/s during a walking
cycle of 1.346 s, along the radars line of sight in an uncluttered environment, is shown for 1 GHz and
5 GHz in Figures 11(a) and 11(b), respectively. Based on the maximum velocity of 5 m/s, we utilize
500 samples to cover a walking cycle for 5 GHz, and 100 samples for 1 GHz. The cyclical movement of
the different body parts can be observed in the spectrograms, with a significantly better resolution at
5 GHz, which clearly shows the body motion as it goes through the different phases of the walking gait
as was shown in Figure 3. We observe that the responses from the different body parts are not easily
distinguishable for the 1 GHz case, and higher frequencies are preferable to identify human motion in a
Doppler signature.
5. SIMULATION RESULTS FROM THE SCENE
The results of our model for the micro-Doppler signatures from a moving human in a forest environment
are presented in this section. A 5 GHz, z-polarized plane wave propagating along the ~x direction is
assumed incident on the forest, which is centered at the origin and covers an area of 40 40 m2 on the
x-y plane as depicted in Figure 12. The circles in the figure represent the trunk positions, and the two
squares denote the starting points for the human for the two paths simulated. A CW radar antenna
with a half power beamwidth of 43.6 degrees is assumed to receive the scattered fields from the scene.
The shaded area in Figure 12 denotes the region of the forest outside the half power beamwidth of the
radar antenna. A brief description on the system parameters and a link budget calculation for the radar
range are provided in Appendix A for this scenario.
40 m

(30,0)
x

RADAR

(0,0)

40 m
PATH 2

Figure 12. Simulation scenario.

(-10,0)

HP

PATH1

Progress In Electromagnetics Research, Vol. 148, 2014

11

A human of height 1.8 m moves with a constant speed of 0.9 m/s along the specified direct paths
to radar inside the forest. The human is modeled using PEC ellipsoids, created based on the Boulic
model as described in Section 2. The forest is modeled using identical trunks of 3 meters height and
11 cm radius which are distributed uniformly inside the 4040 m2 region. PEC is assumed for the trunk
material properties as well as the ground. A trunk density of 3.8% (#/m2 ) is assumed, which amounts
to 35 trunks within the half power beamwidth of the antenna, with average closest neighbor distance
of 5 m. These forest parameters are chosen to reflect the structure and spatial patterns of the trees in
a typical old-growth forest [43].
The Doppler spectrogram from the scene is dependent on the path of motion with respect to the
radar. To illustrate this effect, we consider two direct walking paths in the forest with different starting
points as depicted in Figure 12. Path 1 starts farther away from the radar, and should be exposed to
more attenuation than Path 2. Also, as a consequence of being in the back of the forest, the human
following Path 1 will be in the forward scattering zone of more trees than the human in Path 2. This is
expected to result in more coupling effects from the trunks. We investigate the different spectrograms
for these paths in Figure 13. The same human starts with the same position and moves at the same
speed for the two cases. The first step is taken with the left foot.
We observe that the motion can be detected for both scenarios, although the spectrograms differ
in nature as seen in the different intensity levels for the left and right feet in Figure 13(a), while they
are similar for Path 2 in Figure 13(b). The weaker signals for the left foot compared to the right foot in
Figure 13(a) are due to the coupling effects experienced by the human in Path 1. We also observe that
the expected attenuation effects are visible, as the signature from the path closer to the radar (Path 2)
is 20 dB higher than that of Path 1. This 20 dB difference corresponds to the round trip path loss for the
10 meter separation between the two paths, based on the 1 dB/m specific attenuation assumed for the
forest at 5 GHz. Another feature we notice is that, despite the attenuation, the micro-Doppler signature
still reveals the human motion 30 m into the forest. Finally as expected, the static contribution from
the scene is strong across the 0 Hz band for both cases, masking the lower speed motions due to the
torso.
We demonstrate in Figure 14, the spectrogram from the same scene for both paths without
-90

100

-110

50

-120

-50

150

-100

0.5
Time (s)

(a)

Figure 13. Spectrogram for Eradar

Doppler (Hz)

Doppler (Hz)

150

-70

-80

100

-90

50

-100

-130

-110

-140

-50
0

-120
0.5
Time (s)

(b)
= Ef -r + Eh-r + Ef -h-r , for (a) Path 1 and (b) Path 2.
-70

-90
150

-80

-100

100

-110

Doppler (Hz)

Doppler (Hz)

150

100

50

50

-120

-130

-140

-50
0

-50

0.5

1
Time (s)

(a)

Figure 14. Spectrogram for Eradar

2.

-90
-100
-110

0.5

-120

Time (s)

(b)
= Ef -r + Eh-r , i.e., no mutual coupling, for (a) Path 1 and (b) Path

12

Garcia-Rubia et al.

accounting for any coupling effects; i.e., neglecting the Ef -h-r term in the total received fields at the
radar. Without the mutual coupling effects, the spectrograms are now similar for the right and the
left feet in both paths. We also observe that the spectrogram for Path 2 without the mutual coupling
(Figure 14(b)) is very similar to the spectrogram with the coupling effects (Figure 13(b)), indicating
that for this path the Ef -h-r term was negligible. This can be explained by the fact that the human was
not in the forward direction of the any of the trees closest to his path for Path 2, unlike Path 1 where
the closest few trees were in the direct forward direction.
6. CONCLUSIONS
A capability to detect, track and monitor different human motions in highly cluttered environments,
such as forests, is important for security and surveillance operations. In this paper we develop a model
which utilizes an enhanced full wave computational technique (MoM-FMM) in combination with a firstorder analytical forest model, including specific attenuation inside the forest and mutual coupling effects
between the forest and the human. The solution of this large-scale electromagnetic scattering problem
has been made feasible through a parallel implementation on a 13-node GPU cluster. We demonstrate
that human bipedal movement is discriminable in spectrograms for a walking human in the forest.
We observe that the contribution from the forest masks the Doppler effect of the low radial velocities,
which could happen depending on the path with respect to the radar or due to slow motion. Based
on the specific attenuation values reported in literature for wooded areas, a CW radar at 5 GHz can
discriminate the human motion as far as 30 m into the forest as observed in the spectrogram generated
by our model. Also we note that the contribution from the coupling effects between human and forest
is only significant if the human path is in the forward direction of the closest trees in the forest.
APPENDIX A. RADAR RANGE ESTIMATION
A brief summary of the range estimate of the system based on the radar equation, including the forest
specific attenuation as shown in Recommendation ITU-R P.833-9 [23] is presented in this Appendix.
The radar equation is used to estimate the maximum operating range of our radar:

1/4
PT GT GR 2c
Rmax =
(A1)
(4)3 SN F (S/Nmin )
where
PT = transmitter output power = 60 dBm,
GT = transmitter antenna (horn) gain = 13.3 dBi, (BW3 dB = 43.6 deg E-plane and H-plane),
GR = receiver antenna (horn) gain = 13.3 dBi,
c = carrier frequency wavelength = 0.06 m,
= human body radar cross-section = 0 dBm2 ,
SN F = system noise floor = 130 dBm,
S/Nmin = minimum signal-to-noise ratio = 10 dB.
Using (A1) with the parameters above, we estimate the maximum operating range of our CW radar
operating at 5 GHz for human detection in free-space to be 5,406 m. For forest attenuation of 80 dB
(typical specific attenuation of 1 dB/m for vertical polarization as shown in Recommendation ITU-R
P.833-9 [23], along a woodland path of 40 m), the estimated maximum operating range becomes 54.3 m.
REFERENCES
1. Mencia-Oliva, B., J. Grajal, O. A. Yeste-Ojeda, G. Rubio-Cidre, and A. Badolato, Low-cost CWLFM radar sensor at 100 GHz, IEEE Transactions on Microwave Theory and Techniques, Vol. 61,
No. 2, 986998, 2013.
2. Sume, A., M. Gustafsson, M. Herberthson, A. Janis, S. Nilsson, J. Rahm, and A. Orbom,
Radar detection of moving targets behind corners, IEEE Transactions on Geoscience and Remote
Sensing, Vol. 49, No. 6, 22592267, 2011.

Progress In Electromagnetics Research, Vol. 148, 2014

13

3. Tahmoush, D. and J. Silvious, Remote detection of humans and animals, 2009 IEEE Applied
Imagery Pattern Recognition Workshop (AIPRW), 18, 2009.
4. Otero, M., Application of a continuous wave radar for human gait recognition, Signal Processing,
Sensor Fusion, and Target Recognition XIV, Vol. 5809, 538, Orlando, Florida, USA, Mar. 28, 2005.
5. Chen, V. C., F. Li, S.-S. Ho, and H. Wechsler, Micro-Doppler effect in radar: Phenomenon, model,
and simulation study, IEEE Transactions on Aerospace and Electronic Systems, Vol. 42, No. 1,
221, 2006.
6. Jung, J. H., U. Lee, S. H. Kim, and S. H. Park, Micro-Doppler analysis of Korean offshore wind
turbine on the L-band radar, Progress In Electromagnetics Research, Vol. 143, 87104, 2013.
7. Park, J. H. and N. H. Myung, Effective reconstruction of the rotation-induced micro-doppler from
a noise-corrupted signature, Progress In Electromagnetics Research, Vol. 138, 499518, 2013.
8. Pan, X., W. Wang, J. Liu, D. J. Feng, Y. Liu, and G. Wang, Features extraction of rotationally
symmetric ballistic targets based on micro-doppler, Progress In Electromagnetics Research,
Vol. 137, 727740, 2013.
9. Kim, Y. and H. Ling, Human activity classification based on micro-Doppler signatures using a
support vector machine, IEEE Transactions on Geoscience and Remote Sensing, Vol. 47, No. 5,
13281337, 2009.
10. Fairchild, D. P. and R. M. Narayanan, Classification and modeling of human activities using
empirical mode decomposition with S-band and millimeter-wave micro-Doppler radars, Radar
Sensor Technology XVI, Vol. 8361, Jun. 2012.
11. Lin, A. and H. Ling, Doppler and direction-of-arrival (DDOA) radar for multiple-mover sensing,
IEEE Transactions on Aerospace and Electronic Systems, Vol. 43, No. 4, 14961509, 2007.
12. Li, C., J. Ling, J. Li, and J. Lin, Accurate Doppler radar noncontact vital sign detection using
the RELAX algorithm, IEEE Transactions on Instrumentation and Measurement, Vol. 59, No. 3,
687695, 2010.
13. Zhou, Q., J. Liu, A. Host-Madsen, O. Boric-Lubecke, and V. Lubecke, Detection of multiple
heartbeats using Doppler radar, 2006 IEEE International Conference on Acoustics, Speech and
Signal Processing, ICASSP 2006 Proceedings, Vol. 2, II, 2006.
14. Chen, Y. F., D. Misra, H. Wang, H.-R. Chuang, and E. Postow, An X-band microwave lifedetection system, IEEE Transactions on Biomedical Engineering, Vol. 33, No. 7, 697701, 1986.
15. Chen, Y. F., Y. Huang, J. Zhang, and A. Norman, Microwave life-detection systems for searching
human subjects under earthquake rubble or behind barrier, IEEE Transactions on Biomedical
Engineering, Vol. 47, No. 1, 105114, 2000.
16. Silvious, J., J. Clark, T. Pizzillo, and D. Tahmoush, Micro-Doppler phenomenology of humans at
UHF and Ku-band for biometric characterization, SPIE Defense, Security, and Sensing, 73080X
73080X-9, 2009.
17. Chen, V. C., Detection and analysis of human motion by radar, IEEE Radar Conference,
RADAR 08, 14, 2008.
18. Li, J., X. Zhu, P. Stoica, and M. Rangaswamy, High resolution angle-doppler imaging for MTI
radar, IEEE Transactions on Aerospace and Electronic Systems, Vol. 46, No. 3, 15441556, 2010.
19. Armstrong, B. and B. S. Holeman, Target tracking with a network of Doppler radars, IEEE
Transactions on Aerospace and Electronic Systems, Vol. 34, No. 1, 3348, 1998.
20. Thayaparan, T., L. Stankovi, and I. Djurovi, Micro-Doppler-based target detection and feature
extraction in indoor and outdoor environments, Journal of the Franklin Institute, Vol. 345, No. 6,
700722, 2008.
21. Geisheimer, J. L., W. S. Marshall, and E. Greneker, A continuous-wave (CW) radar for gait
analysis, Conference Record of the Thirty-Fifth Asilomar Conference on Signals, Systems and
Computers, Vol. 1, 834838, 2001.
22. Mobasseri, B. G. and M. G. Amin, A time-frequency classifier for human gait recognition, Proc.
of SPIE, Vol. 7306, 730628, 2009.
23. ITU-R, Attenuation in vegetation, Recommendation ITU-R, 833-8, ITU Radiocommunication

14

24.
25.
26.
27.
28.
29.
30.
31.

32.
33.
34.
35.
36.
37.
38.
39.
40.
41.
42.
43.

Garcia-Rubia et al.

Bureau, Geneva, Switzerland, 2013.

M
uller, M., T. Roder, M. Clausen, B. Eberhardt, B. Kr
uger, and A. Weber, Documentation mocap
database HDM05, Technical Report No. CG-2007-2, ISSN 1610-8892, Universit
at Bonn, 2007.
Boulic, R., N. Thalmann, and D. Thalmann, A global human walking model with real-time
kinematic personification, The Visual Computer, Vol. 6, No. 6, 344358, 1990.
Chen, V. C., Doppler signatures of radar backscattering from objects with micro-motions, IET
Signal Processing, Vol. 2, No. 3, 291300, 2008.
Ram, S. S. and H. Ling, Analysis of microDopplers from human gait using reassigned joint timefrequency transform, Electronics Letters, Vol. 43, No. 23, 2007.
Van Dorp, P. and F. C. A. Groen, Human walking estimation with radar, IEE Proceedings
Radar, Sonar and Navigation, Vol. 150, No. 5, 356365, 2003.
Picard, G. and T. L. Toan, A multiple scattering model for C-band backscatter of wheat canopies,
Journal of Electromagnetic Waves and Applications, Vol. 16, No. 10, 14471466, 2002.
Lang, R. H. and J. S. Sighu, Electromagnetic backscattering from a layer of vegetation: A discrete
approach, IEEE Transactions on Geoscience and Remote Sensing, Vol. 21, No. 1, 6271, 1983.
Tsang, L., K. Ding, G. Zhang, C. C. Hsu, and J. Kong, Backscattering enhancement and clustering
effects of randomly distributed dielectric cylinders overlying a dielectric half space based on MonteCarlo simulations, IEEE Transactions on Antennas and Propagation, Vol. 43, No. 5, 488499,
1995.
Tsang, L., J. A. Kong, K. Ding, and C. O. Ao, Scattering of Electromagnetic Waves, Numerical
Simulations, John Wiley & Sons, 2004.
Ram, S. S., C. Christianson, Y. Kim, and H. Ling, Simulation and analysis of human microDopplers in through-wall environments, IEEE Transactions on Geoscience and Remote Sensing,
Vol. 48, No. 4, 20152023, 2010.
Vahidpour, M. and K. Sarabandi, Millimeter-wave Doppler spectrum and polarimetric response of
walking bodies, IEEE Transactions on Geoscience and Remote Sensing, Vol. 50, No. 7, 28662879,
2012.
Dogaru, T. and C. Le, Time-frequency analysis of a moving human doppler signature, ARL-TR4728, US Army Research Laboratory, Adelphi, MD, 2009.
Tavlove, A. and S. C. Hagness, Computational Electrodynamics: The Finite-difference Timedomain Method, 3rd Edition, Artech House, 2005.
Coifman, R., V. Rokhlin, and S. Wandzura, The fast multipole method for the wave equation: A
pedestrian prescription, IEEE Antennas and Propagation Magazine, Vol. 35, No. 3, 712, 1993.
Nguyen, Q., V. Dang, and O. Kilic, Graphics processing unit accelerated fast multipole method
Fast Fourier transform, IEEE Antennas and Propagation Society International Symposium
(APSURSI 2013), 18821883, Lake Buena Vista, FL, USA, Jul. 712, 2013.
Nguyen, Q. M., V. Dang, O. Kilic, and E. El-Araby, Parallelizing fast multipole method for largescale electromagnetic problems using GPU clusters, IEEE Antennas and Wireless Propagation
Letters, Vol. 12, 868871, 2013.
Rao, S. M., D. Wilton, and A. W. Glisson, Electromagnetic scattering by surfaces of arbitrary
shape, IEEE Transactions on Antennas and Propagation, Vol. 30, No. 3, 409418, 1982.
Leat, C. J., N. V. Shuley, and G. F. Stickley, Triangular-patch model of bowtie antennas:
Validation against Brown and Woodward, IEE Proceedings Microwaves, Antennas and
Propagation, Vol. 145, No. 6, 465470, 1998.
Makarov, S., Antenna and EM Modeling with MATLAB, Princeton University Press, 2002.
Chokkalingam, U. and A. White, Structure and spatial patterns of trees in old-growth northern
hardwood and mixed forests of northern Maine, Plant Ecology, Vol. 156, No. 2, 139160, 2001.