You are on page 1of 6

Radar based detection and tracking of a walking human

Juhana Ahtiainen*, Sami Terho*,


Sampsa Koponen**

*Department of Automation and Systems Technology, **Department of Radio Science and Engineering
Aalto University School of Science and Technology, Espoo, Finland, (e-mail: firstname.lastname@tkk.fi)

Abstract: In this paper, a radar based approach for pedestrian detection and tracking is presented. The
approach utilizes automotive radar for tracking the targets and a separate Doppler radar to provide extra
information about the classes of the targets. The classification of objects as humans is based on analysis
of the Doppler spectrum of targets as well as the speed of the targets gained through the tracking process.
The presented method is able to make a difference between a non-human target and a human target if the
human target is moving. At the moment the measurement platform needs to be stationary or move at very
low speed.
Keywords: Human detection, tracking, automotive radar, Doppler radar, spectrum analysis, filtering

relatively slow speed. If the sensor system is moving, the


1. INTRODUCTION
movements of the platform must be passed to the algorithm.
Human detection and tracking have been widely researched
topics in recent years. Radars are considered suitable sensors 2. APPROACHES TO HUMAN DETECTION WITH
for detecting and tracking objects. However, the classification RADARS
of the objects as humans with radar only, is a very
Classifying targets as humans with radars is a challenging
challenging problem. Therefore, most of the studies in the
task. There are, however, at least three different kind of
field of human detection have been based on other sensors,
techniques applied to this problem. Yarovoy et al. (2008)
such as cameras and laser scanners.
measured the reflectivity of a human in laboratory conditions
Detecting humans with any kind of sensor system is a with ultra wide band (UWB) radar. They analyzed the
challenging task due to the wide variety of positions and polarization of the reflected signal and discovered that there
appearances which humans can assume. A moving sensor are some frequencies where there is a maximum reflectivity
system and stationary people further complicate the task. in one polarization and a minimum in the other. The
Also, the tracking of humans is especially hard since people’s polarized signal, however, depends strongly on the shape,
behaviors are often completely unpredictable. When dealing posture and position of the person making it therefore very
with these problems outdoors, the adverse weather conditions unreliable method for human classification.
also present major limitations to the sensor system used.
Nakane and Soshi (1998 & 2000) used dual band frequency
Certain radars provide reliable measurements regardless of
modulated continuous wave (FMCW) radar for
the weather conditions, hence making radars a desirable
discriminating whether or not the detected object is an
option when considering a human detection system.
animal, including a human. They compared the difference in
In the following, a radar based measurement system for the power of the signal reflected form an object at different
detecting and tracking walking humans is presented. The frequencies (10 GHz and 60 GHz) and were able to establish
measurement system comprises of off-the-shelf automotive a threshold for the ratio of received intensity between two
radar and a continuous wave Doppler radar. The automotive frequencies, above which the detected objects can be
radar is used to track targets and measure their velocities. identified as animals. They received a patent for their method
Doppler radar is used to analyse whether or not there is a in 2000.
human in the measurement sector. Information from both
Otero (2005) analyzed the Doppler spectrum of CW radar to
radars is fused to achieve estimate whether a target is a
obtain a Doppler signature for a walking human. The
human or not. Two different test cases are analyzes with and
movement of a walking human’s legs, arms and torso cause
without the Doppler data revealing the advantages of using
different components to the Doppler frequency. The motion
the Doppler radar. The measurement system is able to track
of the torso produces a steady Doppler shifted signal, which
and detect walking humans. Adverse weather conditions
is modulated by the motion of swinging arms and legs that
deteriorate the classification of observed targets as humans
results in a Doppler signature that is very characteristic of
but do not affect the performance of the tracker. At the
moment, the sensor system needs to be stationary or move at
humans. Otero developed a simple, binary classifier that 3.2 Doppler Radar
classifies detection as either a person present or not present.
The Doppler radar used in the system is Kustom Signals HR-
Thayparan et al. (2004) used wavelet transform to extract the 12 (Kustom Signals 2009). It is speed control radar used by
micro signatures created by human walking. Gürbüz et al. police in some countries. The radar operates on K-band. The
(2007) evaluated different methods used to detect walking exact beam width is not known, but it is very narrow, just a
humans from the Doppler data with the conclusion that few degrees. The intermediate Doppler frequency signal can
spectrogram based methods are problematic if one needs to be acquired from the radar with a special connection. This
distinguish human from other periodically moving targets raw signal is sent to a computer for further signal processing.
due to high signal-to-noise ratio.
Most of the mentioned studies concentrated on identifying 4. ALGORITHM FOR DETECTING A WALKING
the human and distinguishing it from other periodically HUMAN FROM DOPPLER SPECTRUM
moving targets, such as walking animals. In the presence of Detection of the periodic motion can be done by analyzing
other moving targets in the radar’s field of view, this may be the Fourier spectrogram of the Doppler frequency. Fig. 1
impossible because many different spectral signatures are shows a typical spectrogram caused by a walking human. A
superimposed together, and the information from one signal typical feature of this type Doppler signature is that the
is not enough to distinguish them from each other. However, bandwidth of the signal varies periodically. A straightforward
there are situations where it is enough to detect any walking way to measure this bandwidth is to calculate the variance of
object, but it is not necessary to distinguish animals from the signal. However, there may be other frequency
humans. In such cases it is enough to detect just the periodic components caused by other moving objects, such as cars or
motion caused by walking. machines. Therefore, it is necessary to analyze small bands of
the frequency spectrum individually.
3. SYSTEM DESCRIPTION
In this research, three slices of the spectrogram are analyzed.
The measurement system consists of a off-the-shelf The slices are marked to the spectrogram in Fig. 1. Six one-
automotive FMCW radar used in adaptive cruise control dimensional signals are obtained by calculating the variance
systems (ACC) and of a CW Doppler radar used by the police and mean of the frequencies as a function of time. The
to catch speeding violators. The characteristics of these radars signals are shown in Fig. 2. The cadence frequency caused by
are explained in the following chapters. the periodic movement is clearly visible in these signals. The
frequency can be determined with spectral or cepstral
3.1 Automotive Radar analysis, autocorrelation, or auto-difference function, that is,
a sum of squared differences of the signal with itself with
The automotive radar in the measurement system, a long different offsets. In this research, the latter method was used,
range radar 77 GHz generation 2 produced by Bosch, is because it was found to be the most reliable.
described in this chapter. Originally it is manufactured for
ACC systems to measure the distance and relative radial The auto-difference function Dn of the function Fn at time t
velocity of vehicles ahead. The radar has four antennas with and shift s is calculated with
superimposed beams giving the radar system an angular s+w
2
coverage of about 16 degrees. The signal is transmitted Dn (t , s ) = ∑w (Fn (t ) − Fn (t − d ))
2
, (1)
through each of 4 antennas simultaneously. d =s−
2

The radar transmits a continuous wave signal which


where w is the width of the analyzation window. The
frequency is modulated as a function of time with a periodic
Euclidean sum of squares
wave form. The periodic wave consists of three linear
frequency ramps with different ramp times and bandwidth. 6
D(t , s ) = ∑ Dn (t , s )
2
Using three ramps enables the radar system to detect the (2)
range and relative radial velocity of multiple targets at every n =1

time step. Angular position is determined by applying


comparisons of amplitude and phase between adjacent is calculated to form one function D(t,s).
antenna beams (Troppman and Höger, 2008). D(t,s) needs to be low to indicate strong correlation with the
The received signals are processed in the radar shift s. To distinguish low values caused by strong correlation
microcontroller and a complete target list is passed to the from just low signal strength, the values of D need to be
CAN-bus of the radar. Maximum of 32 strongest reflections normalized. A typical characteristics of a periodic signal is
are considered as targets. The following information is that where the value D(t,s) is low, the value D(t,s/2)
presented in the target list from every target: Distance, lateral representing the half shift is high. Therefore the
distance from center line and its variance, relative radial normalization can be done with
speed and relative radial acceleration. Also, time stamps of D(t , s )
D Norm (t , s ) =
each measurement is provided as well as some radar
diagnostics. D t, s( )
2 . (3)
The cycle length of the periodic motion can now determined their own and are not addressed in this paper in more detail.
by finding the local minima of the function DNorm(t,s) in s Interested reader should look up e.g. (Bar-Shalom and
direction. Because the pace of the walking is generally quite Fortman 1998) and (Waltz and Llinas 1990).
slow, the function needs to be observed for at least two
seconds before a walking human can be detected reliably. 5.1 Filtering the measurements of the automotive radar
Fig. 3 shows the output of the DNorm(t,s) function.
Filtering out the clutter measurements is important in order to
reduce false detections. Human movements are modelled
with continuous Wiener process acceleration model (Bar-
Shalom et al. 2001). The process noise – changes in the
acceleration in this case – is modeled as continuous zero-
mean white noise. The measurement noise is also assumed to
be white. Since both dynamic model and measurement model
are linear, Kalman filtering (Kalman 1960) is an obvious
choice for the filtering problem.
Because there are possibly several targets in the field of view
at once, the data association problem needs to be solved. In
this case, a simple Nearest Neighbour (NN) method is used
(Blackman and Popoli 1999). This method associates the
Fig. 1. An excerpt of spectrogram containing measurement of nearest measurement within the validation region to the
a walking human. The three numbered bands are analyzed target. If no target is associated to a measurement, a tentative
individually. target is formed. If another measurement is associated to that
tentative target in the following time step, the target is
confirmed. Targets are deleted if there are no measurements
associated to them in several consecutive time steps. For the
validation region, the uncertainty ellipses of the
measurements and the targets are used. After the data
association is made, separate Kalman filters may be applied
to each target. The NN data association works fine since the
measurements are received with high enough frequency
(10 Hz) and they are not too close with one other. Therefore,
the usage of more complex data association algorithms is not
justified.
Each target is given an extra state – a human probability -
which is updated separately from the Kalman filter. The
human probability ranges from zero to one, one being that it
is absolutely certain that the target is human. If there is no
knowledge of the class of the target, a neutral value of 0.5 is
Fig. 2. Mean (blue curve) and standard deviation (red curve)
assigned to the human probability.
of frequencies as functions of time. The curves are
normalized to same scale. The velocity of each target is monitored and if it is close to
the average walking speed of a human, 1.35 m/s (Anderson
and Pandy 2001), the human probability is increased slightly.
If the velocity is not within the specified interval, the human
probability is decreased moderately. In this research, a
probability value 0.55 was used for targets moving with right
speed, and 0.45 for other targets. Probability value 0.9 was
used for Doppler radar detections. Targets with probabilities
over 0.9 were considered as humans. These values were
Fig. 3. Function DNorm(t,s), with t as time and s as shift. found experimentally.
Darker values represent low values and of the function. The
yellow line indicates the local minima with low enough These probability values are combined at every time step
values. This represents the detected periodicity in the signal. with Bayes rule
p( y | x) p( x) (4)
5. TARGET TRACKING AND DATA FUSION p(x | y) =
p ( y | x ) p ( x ) + (1 − p ( y | x ))(1 − p ( x ))
In this chapter a simple multitarget tracking method and data where y is the measurement and x is the state.
fusion process used in the measurement system are presented.
Both, target tracking and data fusion, are research fields of
If a target travelling close to the average human walking represents the human at the back. As can be seen in the Fig 5,
speed is tracked for a while, the human probability increases there is no target between these humans, hence making the
every time step eventually enabling the classification of that third measurement a false detection.
target as human.

5.2 Data fusion

Fusing data from both radars is a simple procedure, since the


data from the Doppler radar only provides additional
information about the class of the target. It is known that if
there is an observation with the Doppler radar, there is also a
target which is moving roughly at the average walking speed
of a human. Therefore, the target velocity is taken into
consideration at this point as well.
The Doppler measurements are modelled as Gaussian
distributions roughly covering the measurement sector of the
radar. An example of this kind of distribution can be seen in
the Fig. 4 marked with blue dashed line. Data association
between the Doppler measurement and a tracked target is
made if the distribution of the Doppler measurement, at
certain time step, intersects with the estimated position
distribution of that target. If the velocity estimate of such
target lies on the specified human walking speed interval, the Fig. 5. A photograph of the data set gathered from harbour.
human probability of the Doppler measurement is combined Two human targets are walking in the field of view of the
to the human probability of the target with Bayes rule. This sensor system.
way, there can be stationary targets in the area of that
relatively large distribution of the Doppler measurement 6. EXPERIMENTS
which human probabilities will not get falsely updated.
This chapter explains where and how the data sets used to
The fusion process can be clarified with the help of Fig. 4. analyze the method presented in this paper are gathered.
The green ellipses in the figure represent the 2-sigma Also, the performance of the method is presented here in a
uncertainty ellipses of the positions of three targets tracked at way that the advantages of the use of the Doppler radar can
that time. The red dots represent positions of targets be seen.
classified as humans and the small black circle inside the
biggest green ellipse represents a clutter measurement which Four data sets were collected, one inside a mine, one in a
is rather uncertain. The Doppler measurement, illustrated harbor, one with simulated rain, and one where the
with blue dashed line, is associated to all of these three measurement platform was moving. The latter two caused
targets. However, only the red targets are moving at problems to the Doppler detection. In case of rain, the
appropriate velocity and therefore only their human numerous moving rain drops caused strong signal that
probabilities are updated. masked the reflection from a human. In case of moving radar,
the similar effect was caused by the environment that was
moving relative to the radar. Strong radar targets again
masked the signal of the human, and the detection could not
be performed. Therefore, only static cases without rain were
considered.

6.1 Measurements

The two data sets analysed here are collected from a mine
and from a harbor with same measurement platform. Radars
were attached to an ATV enabling easy mobility of the
sensors. In addition, a stereo camera was attached to the
Fig. 4. Visualization of the fused data. Red dots represent measurement platform to produce image information from
targets classified as humans. the tests. The radar measurements were validated based on
the stereo camera data. The movements of the measurement
platform were recorded as well. Fig 6 shows a still picture
A photograph of the situation illustrated in the Fig 4 can be from the data set recorded in the mine. Similar image from
seen in the Fig 5. The left hand red dot in Fig 4 represents the the other data set is presented in Fig 5.
human up front in Fig 5 and the red dot on the right hand side
negative observations. All the information in the Table 1 is
also included in the Fig. 7 and Fig. 8.

Fig. 7. The trajectory of the human target can be seen on the


figure. Black dots represent positions where the human was
Fig. 6. A photograph from the test sequence in the mine. observed but was not classified as human. The red dots are
There is only one human target in the field of view. The walls observations which are classified as humans. The gray dots
of the tunnel produce plenty of reflection back to the represent positions where the human target was not observed
automotive radar. with the radar measurement system. Blue dots are radar
measurements originated from clutter or from some other
Both radars were attached to the same laptop computer, target.
automotive radar via USB-CAN adapter and the Doppler
radar via USB port attachable A/D converter. Since both
sensors were connected to the same computer, data was
automatically synchronized. The gathered data sets were
approximately 30 seconds in length.
In the data set from the mine, there was only one human
target walking in the field of view. The trajectory of that
human can be seen in Fig 7. The human started walking away
from the radar on the left hand side of the tunnel making a
turn back just after the 15 meter marker returning on the right
hand side of the tunnel.
The data set recorded at the harbor features two human
targets in the field of view at once. Trajectories of these
humans are illustrated in Fig 8. The other human walked Fig. 8. Trajectories of two humans in the harbour. The colour
across the measurement sector once starting from the left in codes follow the explanations in the Fig 7.
the vicinity of the 7 meter marker ending up to the right on
the five meter marker. The other human started from the right In the mine, there were a lot of reflections from the walls of
hand side on the 5 meter marker and walked away from field the tunnel. These measurements are not considered as clutter
of view on the left hand side just after the 20 meter marker. since they are caused by real objects. Therefore these
The same human returned to the FOV on the left hand side measurements are excluded from the third line of the table.
approximately on the 11 meter marker and walked out of the However, the exclusion is somewhat subjective since there is
measurement sector on the right around the 3 meter marker. no exact method to tell which measurement originated from
the walls and which did not. In the harbor there was one
6.2 Results systematic false target which caused all the clutter
measurement.
Results of the measurement are summarized in Table 1. The
first row contains the combined number of true positive and The Doppler radar only contributes to the classification of the
false negative observations. The second row contains the total targets. Hence, the advantages of the use of the Doppler radar
number of true positive observations. The third row contains can be seen when comparing the fourth and the fifth line of
the number of observations made due to the clutter. The the table 1. In the mine, the effect of Doppler radar is that 20
fourth and fifth rows contain the total number observations more observations are classified as humans. At the harbor, 75
classified as humans with only the automotive radar and with targets are classified as humans due to the Doppler radar. No
the combination of the automotive radar and the Doppler false human classifications were made in these data sets.
radar respectively. The percentages are calculated with
respect to the combined value of true positive and false
Table 1. Results
ACKNOWLEDGEMENTS
Mine Harbor
This research has been carried out under funding from Tekes
Targets present 89 95 (Finnish Funding Agency for Technology and Innovation),
and FIMA (Forum for Intelligent Machines ry).
Targets detected 80 79
REFERENCES
(89.89 %) (83.16 %)
Clutter targets 15 30 Anderson, F.C. and Pandy, M.G. (2001). Dynamic
(16.85 %) (31.58 %) optimization of human walking. Journal of
Targets classified 29 0 Biomechanical Engineering, 123:381-390.
Bosch (32.58 %) (0 %) Bar-Shalom, Y. and Fortman, T.E. (1988), Tracking and
Data Association, New York: Academic Press.
Targets classified 49 75
Bar-Shalom, Y., Li, X.R. and Kirubarajan, T. (2001).
Bosch + Doppler (55.06 %) (78.95 %)
Estimation with applications to tracking and navigation,
New York: John Wiley & Sons Inc.
Blackman S. and Popoli R. (1999). Design and Analysis of
7. CONCLUSIONS AND FUTURE WORK Modern Tracking Systems, Norwood, MA: Artech
House, 1230 pp. ISBN- 978-1580530064.
The combination of automotive and Doppler radars is Gürbüz, S.Z., Melvin, W.L. and Williams, D.B. (2007).
promising. The inability of an automotive radar to distinguish Detection and identification of human targets in radar
humans from other targets was compensated with Doppler data. SPIE Defense and Security Symposium, Orlando,
radar based detection. Since the beam width of the Doppler FL, USA, April 2007
radar was successfully approximated with Gaussian Kalman, R.E. (1960). A New approach to linear filtering and
distribution, the Kalman filter based multitarget tracking prediction problem. Transaction of ASME, Journal of
method was suitable for the sensor fusion with minor Basic Engineering, 82. pp. 35–45.
modifications. Kustom Signals (2009).
http://www.kustomsignals.com/product_body2.asp?prod
Some aspects of the Doppler radar limit the applications of uct_id=34&cat_id=7, Accessed 30.12.2009
the proposed system. Doppler radar cannot detect humans Nakane, T. and Soshi, K. (1998). A study on recognition of
who are stationary or walking across the radar’s field of view. pedestrians using dual band radar. ITS World
The radar can only detect the motion components that are Conference, Seoul, South Korea, October 1998.
directed towards to or away from the radar. In addition, Nakane, T. and Soshi, K. (2000). Animal body detection
Doppler radar is sensitive to rain and movement of the radar system utilizing electromagnetic waves. US Patent
itself. However, if the environment does not contain any 6072173, Available:
strong radar reflectors, the system can also be used when http://www.patentstorm.us/patents/6072173/description.
moving. The relative speeds of walking humans change, but html
because the proposed algorithm only observes the periodicity Otero, M. (2005). Application of a continuous wave radar for
of the motion, the walking can still be detected. human gait recognition. Technical paper, The MITRE
In presence of other moving targets, they can be falsely Corporation. Available:
classified as humans. This can be a result of walking-like http://www.mitrecorporation.com/work/tech_papers/tech
motion. In the test system, the measurement sectors were _papers_05/05_0330/05_0330.pdf
very narrow for human detection application. A rotating Thayparan, T., Abrol, S., Riseborough, E. (2004), Micro-
measurement system or multiple systems could be a solution. Doppler radar signatures for intelligent target
recognition. Technical paper, Defence R&D Canada.
Fundamentally FMCW and Doppler radars use the same Available:
phenomenon, measuring the frequency shift between http://www.ottawa.drdc-rddc.gc.ca/docs/e/TM-2004-
transmitted and received signal. The automotive radar used 170-eng.pdf
measures also the Doppler frequency. Therefore, if the raw Troppmann, R. and Höger, A. (2008). An introduction to
sensor data of the radar could be accessed, it might be adabtive cruise control (ACC). Driver assistance
possible to implement the algorithms presented in this paper systems, Embedded design India. 4 pp.
using only the data from the automotive radar. The Waltz, E. and Llinas, J. (1990). Multisensor data fusion,
combination of FMCW range and Doppler measurements Boston, MA: Artech house.
would also enable the system to distinguish Doppler Yarovoy, A.G., Ligthart, L.P., Matuzas, J. and Levitas, B.
signatures caused by different people. The Doppler signatures (2008). UWB radar for human being detection. IEEE
would be centered on the frequency respective to the distance A&E Systems Magazine, volume (28), pp. 36-40.
measured with FMCW principle.

You might also like