CC BY
54
MICRO-DOPPLER FEATURE EXTRACTION OF PEDESTRIAN AND AUTOMOBILE USING TWO-DIMENSIONAL FAST FOURIER TRANSFORM
DOI 10.24411/2072-8735-2018-10273
Andrey V. Pluchevskiy,
JSC "Cognitive", Moscow, Russia; Tomsk State University of Control Systems
and Radioelectronics (TUSUR), Tomsk, Russia Keywords: micro-Doppler, Fourier transform, 2D FFT,
[email protected] pedestrian feature extraction, automotive radar.
The current development of microwave and semiconductor technologies provides an ability to use 24 GHz and higher carrier frequencies in radar devices. The higher frequencies provide not only better characteristics but also allow to measure effects that are unavailable on previous frequencies due to low resolution. Thus, it is possible to improve characteristics and improve the abilities of automotive radar systems by measuring, analyzing and processing such effects.
This paper represents the analysis method of a radar signal modulated due to the micro-Doppler effect in unmanned vehicle systems. The analysis shows the connection between the physical movement parameters of the road scene objects and the result of the two-dimensional Fourier transform of the micro-Doppler signal. A pedestrian and an automobile are treated as the road scene objects. The uniform and the accelerated type of movement are studied. The proposed technique is based on vertical and horizontal frequency estimation of a micro-Doppler signal periodic structure. The essential feature of a pedestrian that distinguishes them from an automobile is the vertical periodic components in the micro-Doppler signal. These components implicitly depend on pedestrian velocity and are determined by arm and leg movements. The experiment results verify that the proposed technique corresponds to a theoretical description. The proposed technique can be used in unmanned vehicles and the automotive active safety field. The method is suitable for computer vision systems and is intended for the design of radar automatic target recognition systems.
Information about author:
Andrey V. Pluchevskiy, Junior development engineer, radiolocation department, JSC "Cognitive", Moscow, Russia graduate student, assistant, Tomsk State University of Control Systems and Radioelectronics (TUSUR), Tomsk, Russia
Для цитирования:
Плучевский А.В. Выделение различий между пешеходом и автомобилем основанное на применении двухмерного дискретного преобразования Фурье для анализа сигнала микро-Доплера // T-Comm: Телекоммуникации и транспорт. 2019. Том 13. №5. С. 61-68.
For citation:
Pluchevskiy A.V. (2019). Micro-Doppler feature extraction of pedestrian and automobile using two-dimensional fast Fourier transform. T-Comm, vol. 13, no.5, pр. 61-68.
Introduction
The development of the automotive industry has been rc-eently discussed in many scientific and technical papers \ 1 j. In particular, numerous articles are devoted to computer vision technologies, which are necessary lor unmanned Vehicle control. Computer vision is based on different approaches and different sensors, correspondingly: cameras, radars, lidars, and sonars [2]. An important advantage of radars over other sensors is that the radar characteristics are far less affected by inclement weather conditions such as rain, snow or fog.
The current development of microwave and semiconductor technologies provides an ability to use 24 GHz and higher carrier frequencies in radar devices. The higher frequencies provide not only better characteristics but also allow to measure effects that are unavailable on previous frequencies due to low resolution. Thus, it is possible to improve characteristics and improve the abilities of automotive radar systems by measuring, analyzing and processing such effects.
One of the most important tasks of computer vision systems is the automatic recognition of the road scene and objects inside it. Recognition of pedestrians as a high threat target is of particular importance. The Radar Cross Section (RCS) signatures. Hoppier signatures, and polarization signatures are used for recognition. RCS signatures fluctuate widely due to multipath propagation, especially in the road scene, where at the same lime there occasionally may be a number of objects of a random form and material.
The fluctuation of such signatures can reach 20 dB [3]. Polarization signatures require a special antenna system |4J that leads to an increase in the cost of a radar and a signal process complication. Signatures based on the Doppler effect are more stable in comparison with RCS signatures [3].
Moreover, due to the micro-Doppler (m-D) effect, the pedestrian Doppler spectrogram has unique features [5], which can be used to distinguish a pedestrian from other objects inside the road scene. The micro-Doppler effect is a modulation of the Doppler frequency induced by vibration, rotation or other translation motions of object parts [6].
Even though there are many publications devoted to research and analysis of a human m-D signal [3,5,6,8,10-30], it is still relevant to develop methods of automatic pedestrian recognition in the road scene.
The approach to a recognition radar system design, as described in [7|. consists of 3 following steps:
1) To define the objects and the scenarios to be recognized;
2) data;
3)
To choose the methods of feature extraction from radar
To choose a decision-making criterion.
The analysis of simplified m-D signal models and the pedestrian m-D signal feature extraction technique using two-dimensional fast Fourier transform (2D FFT) is shown in this paper. This technique can be used both to improve existing recognition techniques and to design new ones.
This paper is organized as follows: Section 1 introduces the m-D effect, its parameters, and the measurement method; Section 2 represents the analysis of simplified m-D signal models of the pedestrian and the automobile; the image-based 2D-FFT feature extraction of the m-D signal is shown in Section 4; Section 5 contains the experiment result.
t. Micro-Doppler Signal
When the radar point target moves uniformly and has a constant radial velocity, the constant Doppler frequency shiff occurs on the returned signal. In case the target is distributed and has a complex structure with different parts that have their own movement characteristics, the returned signal will be shifted additionally. The additional modulation of the returned signal induced by vibration, rotation or other translation movements of target parts is known as the micro-Doppler effect [6].
A Doppler spectrum and spectrogram can be obtained by using the fast-ramp technique for LFMCW radar [8-9]. The spectral components of a Doppler spectrum show Doppler frequency shifts for the corresponding target part. It is possible to estimate individual components induced by each target part under the condition that the velocity resolution is sufficient.
The m-D signal of the pedestrian and the automobile is shown in Figure 1. A Doppler spectrogram, where the m-D effect is represented, is meant by the m-D signal, where the m-D effect is represented. In the picture, the automobile is moving towards the radar while the pedestrian is moving away from the radar. The pedestrian m-D signal representation is at the top of Fig. 1, and the automobile m-D signal representation is at the bottom of it. The horizontal axis represents time, the vertical axis represents the Doppler frequency, and the amplitude is shown in color. The form of the spectrogram is discussed in Section 2.
If nf
Emm Eiiii ™ si* ¡Ml v IT™ il" fl ~ ft aY J m
HpStiii
Fig, 1. The Doppler spectrogram. The pedestrian m-D signal is represented at the top, and the automobile m-D signal is represented at the bottom
To describe a Doppler spectrogram containing an m-D signal, the following characteristics are used; signal energy, average Doppler frequency, total bandwidth. Doppler offset, bandwidth without the m-D, and standard deviation (STD) of the lower and upper-frequency envelope [10,11 ¡.
To obtain this spectrogram, AWRI243Boost radar evolution board |12J was used for measurement. DCAI000EVM [13] was used for high-speed radar data ethemet streaming, recording in order to real-time data process on PG. The following parameters were used for LFMCW measurement: 74 GHz carrier frequency, 4 GHz sweep bandwidth, 5 GHz sample rate of complex point ADC, 256 complex samples per chirp, 40 jis chirp duration, 156 ps chirp repetition interval, 256 chirps per one frame, 40ms total frame duration, 25 frames per second.
For each range bin in a sequence of chirps, the mean value of the samples was calculated and subtracted from the samples. For Doppler spectrum computing, FFT calculated for each range bin of the chirp sequence [8] as it is appropriate for the fast ramp
T-Comm Tom 13. #5-2019
technique was used. FFT for range estimation was not calculated. Therefore, there is no information about a target range, and it is not possible to separate objects on the same range and velocity. However, it is suitable tor laboratory experiments where a number of targets and movement scenarios are defined. The reason for skipping range estimation is reducing the complexity of a signal processing algorithm in order to move to the main purpose, namely, the m-D signal analysis. Nevertheless, detection, clustering, and tracking should be implemented so that the proposed method can be used in a real device.
Although Kigure 1 clearly illustrates tlie difference between the pedestrian and automobile m-D signals, it is still necessary to develop techniques of the m-D feature extraction for the following use in computer vision or radar automatic target recognition (ATR) systems.
2. Micro-Doppler signal models
The models and experimental studies of a m-D signal of different human activities including a human gait m-D signal studies are widely available [10, 14-18]. The models of a cyclist are represented in [16, 19]. A vehicle model which helps to predict an automobile m-D signal is shown in [20]. ft is worthwhile to notice that [5] contains m-D signal images of all the aforementioned objects: a pedestrian, a cyclist, and an automobile. The experimental data of this paper are consistent with the listed references.
Tlie form of a pedestrian m-D signal is quite sophisticated. Many reflection points, which have different velocity vectors, are involved in reflected signal forming. Therefore, this paper discusses simplified m-D signal models. Although simplifying leads to information loss, the important information still remains and helps to improve the visual perception of the models. The models of a pedestrian m-D signal and an automobile signal are represented below:
Pedestrian m-D model
Simplified pedestrian m-D signal model
Vertical conipoueuts of ¡ess and ¡inns
Tonso component
of change. The centra! point of this component depicts the pedestrian bulk velocity vp.
The vertical blue bars or vertical components describe the movement of arms and legs. The components amplitude grows in proportion to the pedestrian bulk velocity and changes significantly compared to the body component. The repetition interval decreases as the velocity increases, but not linearly [21]. Legs and arms move mostly simultaneously, that is why the vertical components are fully filled.
Figure 3 shows the representation of the Doppler spectrogram. where a pedestrian moves with acceleration iip. The torso component has the slope 0 in proportion to the pedestrian acceleration. The vertical components do not turn to any direction.
.NJ
X
>> 2 'J
Ci,
Simplified pedestrian m-D signal model
Time, s
Fig. 3. A simplified model of the pedestrian moving ¡¡nearly with acceleration
Automobile m-D model
A uniformly moving automobile m-D signal is shown in Fig. 4. The central component represents the hulk velocity of (he automobile vA. The width of this component stays constant because most of Ihe automobile reflection points have the same velocity vector.
Simplified automobile m-D signal model
■j
ry
p
G. o
Q.
5 C
Car bod y component t'A+VA=2l'A
1 i
1 l'A
1 A VA-1'A=0 It Wheel components
VA
Time, s
Time, s
Fig. 2, The simplified model of the pedestrian moving linearly and uniformly
A horizontal green bar in the center of Fig, 2 represents the Doppler modulation that is induced by torso movement. It is called a horizontal or a torso component of a m-D signal further in the text. This component is characterized by the slow dynamic
Fig. 4. A simplified model of the automobile moving linearly and uniformly
In case the automobile is turned at some angle relative to the radar, addition modulation occurs because the reflection points of a wheel disc have the velocity vector different from the automobile body points. The wheel components can be seen above and below the central component.
The velocities of [he top and bottom points of the wheel are 2vA and 0 m/s, respectively. However, the values of the wheel components will be smaller because tires have less reflection ability than metal discs. Taking into account the height of a tire, the amplitude values of the top and bottom wheel components are
I ,b
= VA ± Vj
R
wheel
where k — frequency, N — number of samples in n direction, - discrete spectrum of function .sjiij. The expression {1 ) can be generalized to 2D case as,
JV-I Ai-1
yjNM n=i) m=<)
, , hi hu
(2)
Where V., is an automobile velocity, is a wheel radius,
and htire is a tire height.
The width of the wheel components depends on a disc form. It is hard to derive the exact formula; however, the general trend is the following: the closer to the radar the component is, the wider it is. Figure J represents this dependency.
The body and top wheel components are rotated at the angle 0 proportional to the automobile acceleration aA, which is shown in Figure 5. The top wheel component is equal to 2vA and turns at a greater angle than the body one, because the body component is equal to vA. The bottom wheel component remains Hat because the velocity of the bottom wheel point stays near zero.
Simplified automobile m-D signal model
where k.l - spatial frequencies in n and m directions, respectively, S[k.!\ - 2D spectrum of 2D function .v|iz,w], N and M are the number of samples in n and m directions, respectively.
FFT along each axis has its own physical meaning. Ordinarily, the first dimension shows the range and the second dimension shows the velocity of the target [8] or the target angle in radar applications. The 3D FFT is used to estimate range, velocity, and angle simultaneously [5,9].
In the previous section, the m-D signal was described as an image. Now let us take a look at the analysis of the images performed with the use of 2D FFT.
The usual view of the ID signal and its spectrum is show n in Figures 6{a) and 6(b), respectively. The 2D signal with a size of NxM is represented in Figure 6(c). The amplitude is shown in color. This 2D signal changes with only one frequency. £1, and along axis n only: thus, there is only one bright point and only on k axis.
1D Spoclrum
Time. S
Fig. 5. A simplified model of the automobile moving linearly with acceleration
To summarize, three main discriminative features of the simplified m-D signal models have been represented: the horizontal periodicity, the vertical periodicity, and the slop. Now an extraction method is required.
3. Feature extraction using 2D FFT
In addition to the direct estimation of spectrogram parameters, different transform methods can be applied, for instance. Discrete Cosine Transform (DCT) [21-23], Pseudo-Zcrciikc moments [24], Cadence Diagram [25], Radon transform [26-27], Mel-frequency cepstral coefficients (MFCC) 123], Log-Gabor filtering [28], Principal Component Analysis (PCA) [29], and Independent Component Analysis (ICA) [30]. Also, two-dimensional Fast-Fourier Transform is appropriate for the feature extraction and this will be shown further.
One-dimension FFT looks like
= (i)
v A H=o
Fig. 6, (a) ID signal with k I frequency: (b) 1D spectrum of the ID signal: (c) 2D signal made of ID signal copies; (d) 2D Spectrum oi'the 2D signal
Signals with an increased frequency relative to Fig, 6 are shown in Figure 7.
The ID signal and spectrum are placed in Fig. 6, 7 to demonstrate the connection between 1D and 2D spectrum. All operations with ID FFT are appropriate with 2D FFT.
Figure 8(a, c) shows signals as images that have a horizontal periodic structure along m axis. The frequencies of these signals are /1, /2, respectively.
The 2D FFT literally decomposes a spatial signal and represents it in a spatial frequency domain. If the signal has a horizontal periodical structure, the 2D spectrum has the frequency point on the horizontal axis as it is illustrated in Figures 6, 7. If the signal has a vertical periodical structure, the 2D spectrum has the frequency point on the vertical axis as il is shown in Fig 8.
The horizontal spatial frequency component in the spectrum in Fig. 11 (b) shows that the m-D signal has a horizontal periodical structure, that is. the periodicity over time. There are vertical spatial frequency components because the signal looks like a pulse in the horizontal or Doppler frequency direction.
The proposed technique is supposed to work in a slide window mode. When a new frame is available, the 2D FFT is recalculated. The minimum window size along the time axis should be not less than 1-1.5s. This time is required to accumulate one period of arms swinging.
Figure 12 shows a selected part of the automobile m-D signal. The mean value of the signal array has also been subtracted.
Automobil□ m-D signal
Car m-D 2D Spectrum
Cycles per &
Fig. 12. (a) Automobile m-D signal (bl 2D spectrum of the automobile m-D signal
This result corresponds to the simplified models described in Section 2.
Conclusion
The image-based analysis of the simplified pedestrian and automobile m-D signal models have been represented in this paper. The m-D signal was described in terms of the vertical structure periodicity and the horizontal structure periodicity that corresponds to the periodicity along the time axis and the Doppler frequency axis. The main feature of a pedestrian which distinguishes them from an automobile is the m-D signal structure that is periodic along the lime axis (Fig. 2),
The suitable method of the feature extraction hased on 2D FFT was also shown. The extraction technique was verified experimentally. Therefore, the FFT can be used not only for estimating physical parameters of targets such as range, velocity or angle but also for the m-D feature extraction. However, different transforms such as DCT can be used instead of FFT,
The advantage of the proposed technique is utilizing the existing hardware abilities of automotive radars and increasing the recognition characteristics of radar systems. It is possible to identify the following disadvantages: it takes much time (1-1.5s) to obtain the first result of extraction and requires tracking and clustering in a multitarget (real) situation. Besides, there is a strong dependence on the radial component of velocity. Thus, the proposed technique is considered as an additional feature extraction method. It can be used to extend possibilities and improve characteristics of the existing recognition systems.
In the luture, the proposed feature extraction method may be used to develop a technique of pedestrian recognition in a road scene. Furthermore, studying the properties of the cyclist and motorcyclist m-D signals is required for developing their feature
extraction techniques in order to establish a complex ATR method.
Acknowledgment
The author thanks C-Pilot team, Manokhin Oleb, Velikanova Elena, Kostarev Aleksey for their contributions and special thanks to Popov Yuriy for help in the research organizing.
References
1. Ilasch. J. (2015). Driving towards 2020: Automotive radar lech no logy trends. The 2015 IEEE MTT-S International Conference on Microwaves for Intelligent Mobility (ICMIM'). pp. ! -4.
2. UnitedLex Blog Posts - From Google to Tesla, it's a War of LiDAR or RADAR (2017). [online] Availible at: https://www.unftedlex.COTn/news/from-gOQgle-to4e5la-it5-a-war-of-lidar-or-radar (Accessed 15 Mar,2019).
3. Bartseli, A., Kitzek, p., & Rasshofer. R. H. (2012). Pedestrian recognition using automotive radar sensors. Advances in Radio Science, Vol.10, no. 2, pp. 45-55.
4. Milligan, T. A. (2005). Modern antenna design, 2ed ed. New Jersey: IEEE press. 614 p.
5. Wagner, T., Feger, R„ & Stelzer, A. (2017). Radar signal processing for jointly estimating tracks and miero-doppler signatures. IEEE Access, Vol. 5, pp. 1220-1238.
6. Chen, V. C„ Li, F., Ho, S. S.. & Weehsler, H. (2006). Micro-Doppler effect in radar: phenomenon, model, and simulation study, IEEE Transactions on Aerospace and electronic systems, Vol. 42, no. I, pp. 2-21.
7. Gorelik A.L., Barabash Ju.L., Krivosheev O.V. (IW0). Selek-cija i raspoznavanfe na osnove lokacionnuj mformacii (Radar information based recognition and selection). Moskva: Radio i svjaz, 240p. {in Russian)
8. Hyun, E„ Jin, Y. S., & Lee, J. H. (2016). A pedestrian detection scheme using a coherent phase difference method based on 2D range-Doppler FMCW radar, Semors. Vol. 16no, 1, pp. 124-137.
9. Hyun, E. (2015). Parallel and pipelined hardware implementation of radar signal processing lor an FMCW multi-channel radar. Elek-tronika ir Elektrotechmku, Vol. 21, no. 2, pp. 65-71.
10. Zenaldin, M., &. Narayanan, R. M. (2016). Radar micro-Doppler based human activity classification for indoor and outdoor environments, Proc. SPIE, [online] vol. 9829, no. 98291. Available at: hrtp;//dx.doi.org/IO.I117/12.2228397 (Accessed 18 Feb.2019)
11. Narayanan, R. M., & Zenaldin, M. (2015). Radar micro-Doppler Signatures of various human activities. IET Radar, Sonar dt Navigation. Vol, 9, no. 9, pp, 1205-1215,
12. AWRI243 76-GHz to 81-GHz high-performance automotive MMIC evaluation module (20I9(. [online] Availible at: http://www.ti.com/tooI/AWR1243BOOST (Accessed 2 Apr.2019)
13. Real-time data-capture adapter for radar sensing evaluation module (2019). [online] Availible at: http://www.ti.com/toott0CA1OOOEVM (Accessed 2 Apr.2019)
14. Chen, V. C„ Miceii, W. J., & Tahmoush, D. (2014). Radar micro-Doppler signatures; processing and applications. London: The Institution of Engineering and Technology. 406p.
15. Belgiovarse, D., & Chen, C. C. (2017). Micro-Doppler characteristics of pedestrians and bicycles for automotive radar sensors at 77 GHz. The 2017 I Ith European Conference on Antennas and Propagation (EOCAP). pp. 2912-2916.
16. Ghaleb, A., Vignaud, L., & Nicolas, J. M. (2008). Micro-Doppler analysis of wheels and pedestrians in ISAR imaging IET Signal Processing, Vol. 2, no. 3, pp. 301-311.
17. Schubert. E.. MeinL F„ Kunert, M„ & Menzel, W (2015). High resolution automotive radar measurements of vulnerable road us-
ers-pedestrians & cyclists. The 201? IEEE MTT-S International Conference on Microwaves for Intelligent Mobility (ICMIM). pp. I -4i
18. Stolz, M-, Schubert, E., Meinl, F., Kunert, M., & Menzel, W. (2017). MuIii-target reflection point model of cyclists for automotive radar. The 2017 European Radar Conference (EURAD). pp. 94-97.
19. Li, Y., Dli. L., & Liu, H. (2D! 1). Moving vehicle classification based on micro-Doppler signature. The 2011 IEEE International Conference on Signal Processing, Communications and Computing
(ICSPCCh pp. 1-4.
20. Ford, M. P., Wagenaar, R, C.. & Newell. K. M. (2007). Ami constraint and walking in healthy adults. Gait & posture. Vol. 26, no !, pp. 135-141.
21. Gürbiiz, S. Z„ Erol, B., Ca|)iyan, B.. &. Tekeli, B. (2015). Operational assessment and adaptive selection of micro-Doppler features. IET Radar. Sonar & Navigation. Vol. 9, no. 9, pp. \ 196-1204.
22. Vishwakarma. S„ & Ram, S. S. (2018). Dictionary learning with low computational complexity for classification of human micro-Dopplers across multiple carrier frequencies. IEEE Access. Vol. 6, pp. 29793-29805.
23. Erol, B„ Amin. M. 0., & Gurbuz, S". Z. (2018). Automatic data-driven frequency-warped cepstral feature design for micro-Doppler classification. IEEE Transactions on Aerospace and Electronic Systems. Vol.54, no. 4, pp. 1724-1738.
24. Pal lot ta, L.. Clemente, C.. De Main. A.. S о rag h an, J.,!., & Farina, A, (2014). Pseudo-Zemike moments based radar micro-Doppler classification. The 2014 IEEE Radar Conference. pp. 850-854.
25. Zhang. W., & U, G. (2018). Detection of multiple micro-drones via cadence velocity diagram analysis. Electronics Letters. Vol. 54, no. 1. pp. 441-443.
26. Stankovic, L., Dakovic, M., Thayaparan, Т., & Popov ic-Bugarin, V. (2015). Inverse radon transform-based micro-Doppler analysis from a reduced set of observations, IEEE Transactions cm Aerospace and Electronic Systems. Vol. 51, no. 2, pp. 1155-1 169.
27. Li, W,, Xiong, В., & K.uang, G. (2017). An enhancing normalized inverse radon transform for parameter Extraction of mtcro-Doppler. The 2017 2nd International Conference on Frontiers of Sensors Technologies (ICFSTh pp. 354-357.
28. Tivive, F. H. C„ Phung. S. L„ & Gouzerdoum, A. (2015). Classification of micro-Doppler signatures of human motions using logGabor fillers. IET Radar, Sonar t6 Navigation. Vol. 9, no. 9. pp. I 188-I 195.
29. Zabalza, J., Clemente, C'., Di Caterina, G., Ren, J., Soraghan, .1. ,).. & Marshall. S. (2014). Robust PCA micro-Doppler classification using SVM on embedded systems. IEEE Transactions on Aerospace and Electronic Systems. Vol. 50, no. 3, pp. 2304-2310.
30. Chen, V. C. (2005). Spatial and temporal independent component analysis of micro-Doppler features. The IEEE Internationa! Radar Conference. pp. 348-353,
ВЫДЕЛЕНИЕ РАЗЛИЧИЙ МЕЖДУ ПЕШЕХОДОМ И АВТОМОБИЛЕМ ОСНОВАННОЕ НА ПРИМЕНЕНИИ ДВУХМЕРНОГО ДИСКРЕТНОГО ПРЕОБРАЗОВАНИЯ ФУРЬЕ ДЛЯ АНАЛИЗА СИГНАЛА МИКРО-ДОПЛЕРА
Плучевский Андрей Владимирович,
AO "КОГНИТИВ", Москва, Россия; Томский государственный университет систем управления и радиоэлектроники (ТУСУР), г. Томск, Россия,
Аннотация
Представлена методика анализа радиолокационного сигнала, имеющего модуляцию микро-Доплера, в системах беспилотного транспорта. Данный анализ устанавливает связь между физическими параметрами движения объектов дорожной сцены и результатом двухмерного преобразования Фурье спектрограммы Доплера. В качестве объектов рассматриваются пешеход и автомобиль. Рассматриваемый характер движения как равномерное, так и равноускоренное. Методика заключается в оценке вертикальной и горизонтальной периодичности двухмерной структуры сигнала микро-Доплера. Показано, что существенным признаком пешехода, который позволяет отличить его от автомобиля, является наличие периодических вертикальных составляющих в сигнале микроДоплера. Эти составляющие косвенно зависят от скорости движения пешехода и обусловлены движением его рук и ног. Результаты экспериментальной апробации показывают, что предлагаемый подход к анализу соответствует теоретическому описанию. Предлагаемый метод можно применять в отрасли беспилотного транспорта и систем активной безопасности. Данный метод подходит для систем компьютерного зрения и предназначен для построения радиолокационных систем автоматического распознавания целей.
Ключевые слова: микро-Доплер, преобразование Фурье, двухмерное быстрое преобразование Фурье, выделение признаков пешехода, автомобильный радиолокатор.
Литература
1. Hasch, J. (2015). Driving towards 2020: Automotive radar technology trends. The 2015 IEEE MTT-S International Conference on Microwaves for Intelligent Mobility (ICMIM), pp. 1-4.
2. UnitedLex Blog Posts - From Google to Tesla, it's a War of LiDAR or RADAR (2017). [online] Availible at: https://www.unitedlex.com/news/from-google-to-tesla-its-a-war-of-lidar-or-radar (Accessed 15 Mar.2019)
3. Bartsch, A., Fitzek, F., & Rasshofer, R. H. (2012). Pedestrian recognition using automotive radar sensors. Advances in Radio Science, Vol. 10, no. 2, pp. 45-55.
4. Milligan, T. A. (2005). Modern antenna design, 2ed ed. New Jersey: IEEE press. 614 p.
5. Wagner, T., Feger, R., & Steizer, A. (2017). Radar signal processing for jointly estimating tracks and micro-doppler signatures. IEEE Access. Vol. 5, pp. 1220-1238.
6. Chen, V.C., Li, F., Ho, S.S., & Wechsier, H. (2006). Micro-Doppler effect in radar: phenomenon, model, and simulation study. IEEE Transactions on Aerospace and electronic systems. Vol. 42, no. 1, pp. 2-21.
7. Горелик А.Л., Барабашh Ю.Д.., Кривошеев O.B. (1990). Селекция и распознавание на основе локационной информации. M.: Радио и связь, 240 с.
8. Hyun, E., Jin, Y.S., & Lee, J.H. (2016). A pedestrian detection scheme using a coherent phase difference method based on 2D range-Doppler FMCW radar. Sensors. Vol. 16 no. 1, pp. 124-137.
9. Hyun, E. (2015). Parallel and pipelined hardware implementation of radar signal processing for an FMCW multi-channel radar. Elektronika ir Elektrotechnika. Vol. 21, no. 2, pp. 65-71.
10. Zenaidin, M., & Narayanan, R.M. (2016). Radar micro-Doppler based human activity classification for indoor and outdoor environments, Proc. SPIE, [online] vol. 9829, no. 98291. Available at: http://dx.doi.org/l0.lll7/l2.2228397 (Accessed 18 Feb.2019)
11. Narayanan, R.M., & Zenaidin, M. (20l5). Radar micro-Doppler signatures of various human activities. IET Radar, Sonar & Navigation. Vol. 9, no. 9, pp. l205-l2l5.
12. AWR1243 76-GHz to 8l-GHz high-performance automotive MMIC evaluation module (2019). [online] Availible at: http://www.ti.com/tool/AWRl243BOOST (Accessed 2 Apr.20l9)
13. Real-time data-capture adapter for radar sensing evaluation module (2019). [online] Availible at: http://www.ti.com/tool/DCAl000EVM (Accessed 2 Apr.20l9)
14. Chen, V. C., Miceii, W. J., & Tahmoush, D. (20l4). Radar micro-Doppier signatures: processing and applications. London: The Institution of Engineering and Technology. 406 p.
15. Beigiovane, D., & Chen, C.C. (20l7). Micro-Doppler characteristics of pedestrians and bicycles for automotive radar sensors at 77 GHz. The 20l7 llth European Conference on Antennas and Propagation (EUCAP), pp. 29l2-29l6.
16. Ghaieb, A., Vignaud, L., & Nicoias, J.M. (2008). Micro-Doppler analysis of wheels and pedestrians in ISAR imaging. IET Signal Processing. Vol. 2, no. 3, pp. 30l-3ll.
17. Schubert, E., Meini, F., Kunert, M., & Menzei, W. (20l5). High resolution automotive radar measurements of vulnerable road users-pedestrians & cyclists. The 20l5 IEEE MTT-S International Conference on Microwaves for Intelligent Mobility (ICMIM), pp. l-4.
18. Stoiz, M., Schubert, E., Meini, F., Kunert, M., & Menzei, W. (20l7). Multi-target reflection point model of cyclists for automotive radar. The 20l7 European Radar Conference (EURAD), pp. 94-97.
19. Li, Y., Du, L., & Liu, H. (201 l). Moving vehicle classification based on micro-Doppler signature. The 201 l IEEE International Conference on Signal Processing, Communications and Computing (ICSPCC), pp. l-4.
20. Ford, M.P., Wagenaar, R.C., & Neweii, K.M. (2007). Arm constraint and walking in healthy adults. Gait & posture. Vol. 26, no. l, pp. l35-l4l.
21. Girbiz, S. Z., Eroi, B., laiiiyan, B., & Tekeii, B. (2015). Operational assessment and adaptive selection of micro-Doppler features. IET Radar, Sonar & Navigation, Vol. 9, no. 9, pp. ll96-l204.
22. Vishwakarma, S., & Ram, S.S. (2018). Dictionary learning with low computational complexity for classification of human micro-Dopplers across multiple carrier frequencies. IEEE Access, Vol. 6, pp. 29793-29805.
23. Eroi, B., Amin, M.G., & Gurbuz, S.Z. (20l8). Automatic data-driven frequency-warped cepstral feature design for micro-Doppler classification. IEEE Transactions on Aerospace and Electronic Systems. Vol. 54, no. 4, pp. l724-l738.
24. Paiiotta, L., Ciemente, C., De Maio, A., Soraghan, J. J., & Farina, A. (20l4). Pseudo-Zernike moments based radar micro-Doppler classification. The 20l4 IEEE Radar Conference, pp. 850-854.
25. Zhang, W., & Li, G. (20l8). Detection of multiple micro-drones via cadence velocity diagram analysis. Eiectronics Letters. Vol. 54, no. 7, pp. 44l-443.
26. Stankovic, L., Dakovic, M., Thayaparan, T., & Popovic-Bugarin, V. (20l5). Inverse radon transform-based micro-Doppler analysis from a reduced set of observations. IEEE Transactions on Aerospace and Electronic Systems, Vol. 5l, no. 2, pp. ll55-ll69.
27. Li, W., Xiong, B., & Kuang, G. (20l7). An enhancing normalized inverse radon transform for parameter extraction of micro-Doppler. The 20l7 2nd International Conference on Frontiers of Sensors Technologies (ICFST), pp. 354-357.
28. Tivive, F. H. C., Phung, S.L., & Bouzerdoum, A. (20l5). Classification of micro-Doppler signatures of human motions using log-Gabor filters. IET Radar, Sonar & Navigation, Vol. 9, no. 9, pp. ll88-ll95.
29. Zabaiza, J., Ciemente, C., Di Caterina, G., Ren, J., Soraghan, J. J., & Marshaii, S. (20l4). Robust PCA micro-Doppler classification using SVM on embedded systems. IEEE Transactions on Aerospace and Electronic Systems, Vol. 50, no. 3, pp. 2304-23l0.
30. Chen, V.C. (2005). Spatial and temporal independent component analysis of micro-Doppler features. The IEEE International Radar Conference, pp. 348-353.
Информация об авторе:
Плучевский Андрей Владимирович, младший инженер разработчик, департамент радиолокации, AO "КОГНИТИВ", Москва, Россия; аспирант, ассистент кафедры телекоммуникаций и основ радиотехники (ТОР), Томский государственный университет систем управления и радиоэлектроники (ТУСУР), г. Томск, Россия
T-Comm ^м 13. #5-2019
