KR102524373B1 - Human-Detecting Radar System for Indoor Security Applications and Human-Detecting method using the same - Google Patents

Abstract

A human body detection radar system for indoor security application according to an embodiment of the present invention includes a signal processing unit receiving a measurement signal of a CW radar and generating a spectrogram through STFT signal processing; a feature extractor extracting four features based on the spectrogram; and an SVM binary classification unit that classifies whether the object is a human using all of the four features.

Description

Human-Detecting Radar System for Indoor Security Applications and Human-Detecting method using the same}

The present invention relates to a human body detection radar system for indoor security applications and a human body detection method using the same.

In modern society, a security and surveillance system that detects an intruder is essential to maintain the stability of a dwelling and security within a specific space. In particular, a low-area, low-power system is required due to the characteristics of a monitoring system that operates 24 hours a day. In existing security systems, various sensors such as CCTV cameras, infrared sensors, microwave sensors, and radar are used.

However, these sensors often fail to detect an object in a dark environment or malfunction by detecting the movement of an object other than a person. In particular, in the case of CCTV, there is a disadvantage that a person must directly monitor the screen [2]. On the other hand, a radar that detects a target using electromagnetic waves has the advantage of being less affected by the external environment and being able to detect the target even in a dark environment. In addition, research on target recognition, which classifies people and other objects by detecting not only objects but also various human motions or minute movements, is being actively conducted.

S. Oh, S. Moon, S. Choi, “Intelligence Security and Surveillance System in Sensor Network Environment Using Integrated Heterogeneous Sensors” Korea Institute Of Communication Sciences, Vol. 38C, no. 07, p. 551-562, Jul. 2013.

An object of the present invention is to provide a human body detection radar system for indoor security applications that can solve the conventional problems and a human body detection method using the same.

여기서, X는 분류할 입력 벡터이고,

는 평균,

는 표준편차,

는 선형 예측 변수,

는 커널 스케일(kernel scale)이고, b는 바이어스이다.A human body detection radar system for indoor application security according to an embodiment of the present invention for solving the above problems receives a radar signal measuring an object from a CW radar and performs a short-time Fourier transform (STFT) signal processing to form a spectrogram. a signal processing unit that generates; a feature extractor extracting four features based on the spectrogram; and an SVM binary classifier for classifying whether the target is a human as 1 or 0 using all of the four features, wherein the feature extractor includes a time interval of micro-Doppler frequencies in the spectrogram, the The features of the difference between the frequency of the body and the peak frequency in the spectrogram are extracted, whether the number of strides of the target has occurred, the reflection area of the CW radar, and the peak frequency is a micro-Doppler frequency is the maximum value of , and the SVM binary classifier performs binary classification through the following equation using a parameter learned to maximize a margin.
[ceremony]

where X is the input vector to be classified,

is the mean,

is the standard deviation,

is a linear predictor,

is the kernel scale, and b is the bias.

여기서, X는 분류할 입력 벡터이고,

는 평균,

는 표준편차,

는 선형 예측 변수,

는 커널 스케일(kernel scale)이고, b는 바이어스이다.In order to solve the above problems, a method of operating a human body detection radar system for indoor security application according to an embodiment of the present invention is to receive a frequency (radar) signal measured by a CW radar in a signal processing unit and STFT (Short-Time A signal processing step of generating a spectrogram through Fourier Transform) signal processing; a feature extraction step of extracting four features based on the spectrogram in a feature extraction unit; and an SVM binary classification step of classifying whether the object is a human as 1 or 0 using all of the four features in an SVM binary classification unit, wherein the feature extraction step includes a micro-Doppler frequency (micro-Doppler frequency) within the spectrogram. A step of extracting a plurality of features based on a time interval of frequency, whether a count of stride of the object has occurred, a radar reflection area, and a difference between a peak frequency and a body frequency, wherein the peak frequency is a fine Doppler frequency (micro-Doppler frequency), and the SVM binary classification step is characterized by maximizing a margin and classifying the learned parameter through the following equation.
[ceremony]

where X is the input vector to be classified,

is the mean,

is the standard deviation,

is a linear predictor,

is the kernel scale, and b is the bias.

Therefore, the human body detection radar system for indoor security applications according to one embodiment and another embodiment of the present invention applies a signal processing using STFT, a feature extraction technique for efficient classification, and a classification algorithm using SVM, It provides an advantage of increasing the accuracy of human body detection while minimizing computational complexity by simplifying complexity.

That is, the human body detection radar system of the present invention extracts the spectrogram including time information by applying STFT from the CW radar and classifies it into SVM, so that the human body can be detected with high accuracy even with a small area.

1 is a schematic configuration diagram of a human body detection radar system for indoor security applications according to one embodiment and another embodiment of the present invention.
FIG. 2 is a block diagram showing detailed configurations of each component shown in FIG. 1 .
3 is an exemplary diagram of a spectrogram by STFT.
4 is an exemplary view of a spectrogram for each target type.
5 is an exemplary diagram of a spectrogram for a stride period according to an object.
6 is an exemplary view of a spectrogram for the number of steps according to an object.
7 is an exemplary view of a spectrogram showing a radar reflection area according to a target.
8 is an exemplary view of a spectrogram displaying a difference between a maximum frequency and a torso frequency according to an object.
9 is a plan view showing the hyperplane and support vectors of the SVM.
10 is a flowchart illustrating a human body detection method according to an embodiment of the present invention.
11 is a flowchart illustrating a human body detection method according to another embodiment of the present invention.
12 is a photograph showing various experimental environments to evaluate the classification performance of the proposed system.
13 is a graph comparing five feature values.
14 is a table showing the average accuracy for each number of features used in binary classification.
15 is a table showing the accuracy of the classification algorithm.
16 is a table showing a summary of resource usage.
17 is a table showing resource usage by block
18 is an illustration of an example computing environment in which one or more embodiments set forth herein may be implemented.

Hereinafter, embodiments of the present specification will be described with reference to the accompanying drawings. However, it should be understood that the technology described herein is not limited to specific embodiments, and includes various modifications, equivalents, and/or alternatives of the embodiments herein. . In connection with the description of the drawings, like reference numerals may be used for like elements. In this specification, expressions such as “has,” “can have,” “includes,” or “can include” indicate the existence of a corresponding feature (eg, numerical value, function, operation, or component such as a part). , which does not preclude the existence of additional features.

In this specification, expressions such as “A or B,” “at least one of A and/and B,” or “one or more of A or/and B” may include all possible combinations of the items listed together. . For example, “A or B,” “at least one of A and B,” or “at least one of A or B” (1) includes at least one A, (2) includes at least one B, Or (3) may refer to all cases including at least one A and at least one B.

Expressions such as "first," "second," "first," or "second," as used herein, may modify various elements regardless of order and/or importance, and may refer to one element as It is used only to distinguish it from other components and does not limit the corresponding components. For example, a first user device and a second user device may represent different user devices regardless of order or importance. For example, a first element may be termed a second element without departing from the scope of rights described herein, and similarly, the second element may also be renamed to the first element.

A component (e.g., a first component) is "(operatively or communicatively) coupled with/to" another component (e.g., a second component); When referred to as "connected to", it should be understood that the certain component may be directly connected to the other component or connected through another component (eg, a third component). On the other hand, when an element (eg, a first element) is referred to as being “directly connected” or “directly connected” to another element (eg, a second element), the element and the above It may be understood that other components (eg, a third component) do not exist between the other components.

As used herein, the expression “configured (or configured) to” means, depending on the circumstances, for example, “suitable for,” “having the capacity to” ," "designed to," "adapted to," "made to," or "capable of." The term "configured (or set) to" may not necessarily mean only "specifically designed to" hardware. Instead, in some contexts, the phrase "device configured to" may mean that the device is "capable of" in conjunction with other devices or components.

For example, the phrase "a processor configured (or configured) to perform A, B, and C" may include a dedicated processor (eg, embedded processor) to perform the operation, or by executing one or more software programs stored in a memory device. , may mean a general-purpose processor (eg, CPU or application processor) capable of performing corresponding operations.

Terms used in this specification are only used to describe a specific embodiment, and may not be intended to limit the scope of other embodiments. Singular expressions may include plural expressions unless the context clearly dictates otherwise. Terms used herein, including technical or scientific terms, may have the same meaning as commonly understood by one of ordinary skill in the art described herein. Among the terms used in this specification, terms defined in a general dictionary may be interpreted as having the same or similar meaning as the meaning in the context of the related art, and unless explicitly defined in the present specification, an ideal or excessively formal meaning. not be interpreted as In some cases, even terms defined in this specification cannot be interpreted to exclude embodiments in this specification.

Hereinafter, based on the accompanying drawings, a human body detection radar system for indoor security applications and a human body detection method using the same according to an embodiment and another embodiment of the present invention will be described in detail.

1 is a simplified configuration diagram of a human body detection radar system for indoor security applications according to an embodiment of the present invention, and FIG. 2 is a detailed configuration diagram of each configuration shown in FIG. 1 .

1 and 2, the human body detection radar system 100 for indoor security application according to an embodiment of the present invention includes a signal processing unit 110, a feature extraction unit 120 and an SVM binary classification unit 130 includes

The signal processor 110 may be configured to receive a frequency (radar) signal measured by the CW radar 10 and generate a spectrogram through Short-Time Fourier Transform (STFT) signal processing.

The frequency (radar) signal is a Doppler frequency signal that is a difference between a transmission frequency and a reception frequency for the target, which is exhibited by the Doppler effect of the CW radar.

The signal processing unit 110 may calculate the velocity of the target detected by the CW radar 10 using the Doppler frequency signal as shown in Equation 1 below.

[Equation 1]

V is the speed at which the target approaches the radar, and c is the speed of the radio wave.

In addition, the signal processing unit 110 divides the radar signal into short-time signals by applying a window function, calculates the Doppler frequency according to the time change for the divided short-time signals by applying the FFT algorithm, and calculates the result value It is possible to generate the spectrogram by arranging on the time axis.

That is, the spectrogram is obtained by using STFT on the radar signal, applying a window function to divide the radar signal into short-time signals, and applying FFT to the divided short-time signals to obtain Doppler according to time change. frequency can be analyzed.

When the FFT results obtained in this way are connected on the time axis, the spectrogram shown in FIG. 3 can be obtained. Here, the horizontal axis of FIG. 3 represents time, the vertical axis represents speed, and the strength of the signal is represented by color. The closer to deep red, the higher the intensity of the signal, and the closer to blue, the lower the signal strength.

Next, the feature extractor 120 may be configured to extract a plurality of features based on the spectrogram.

Prior to describing the feature extraction unit 120, the frequencies in the spectrogram referred to in the present invention are different depending on the subject. For example, the intensity of the frequency of the torso of the object having the largest area is the strongest, and a micro-Doppler frequency appears due to the movement of the limbs of the object.

For example, dogs move their legs faster than humans, so the spectrogram of FIG. 4 (b) shows more dense micro-Doppler time intervals than the human spectrogram of (a) of FIG. 4 . 4(c) is a spectrogram of the robot cleaner without leg movement, and does not show a fine Doppler frequency. Therefore, in the present invention, the features of the object are extracted through four efficient feature extraction methods based on the movement and shape of the object.

More specifically, the feature extractor 120 first extracts a plurality of features based on an average value of time intervals of micro-Doppler frequencies in the spectrogram.

For example, the period of human leg movement is different from that of short-legged dogs. The time it takes from moving one foot to moving the next is longer in long-legged people than in short-legged dogs. This can be confirmed by the micro-Doppler time interval on the spectrogram (see FIG. 5). That is, a plurality of features of the object may be extracted as an average value of time intervals of micro-Dopplers displayed vertically in FIG. 5 .

In addition, the feature extraction unit 120 may extract a plurality of features based on whether or not micro Doppler frequencies occur in the spectrogram and the number of occurrences.

More precisely, a plurality of features of the object may be extracted based on the number of strides of the object.

For example, as shown in FIG. 6 , unlike a human or a dog, a legless robot cleaner does not show micro Doppler caused by leg movements. This is suitable as a feature that distinguishes a person from an object without legs.

Also, the feature extractor 120 may extract a plurality of features using a radar cross section (RCS) within the spectrogram.

For example, radar cross section (RCS) is the total amount of energy that electromagnetic waves are reflected from an object. The reflected area varies depending on the size of the target, and the larger the target, the larger it appears. Therefore, the radar cross section (RCS) appears large in the order of people, dogs, and robot cleaners. Accordingly, a plurality of features of the object may be extracted through a value calculated by adding all data equal to or greater than a threshold value in the spectrogram.

The feature extractor 120 may extract a plurality of features using a difference between a peak frequency and a body frequency in the spectrogram.

Here, the peak frequency is the maximum value of the micro-Doppler frequency and means the swing speed of the leg.

The difference between the speed of the subject's torso and the swing speed of the legs is determined by the length of the subject's legs. Dogs have shorter legs than humans, so the difference between the body speed and the swing speed of the legs is smaller (see Fig. 7).

Accordingly, a plurality of features of the object may be extracted by calculating a difference between a trunk frequency indicating a trunk speed and a peak frequency indicating a leg swing speed on the spectrogram.

Next, the SVM binary classification unit 130 may be configured to classify an object as 1 or 0 using four features related to the movement of the human body.

The SVM binary classification unit 130 classifies features into the two classes using Equation 2 below.

[Equation 2]

는 평균,

는 표준편차,

는 선형 예측 변수,

는 커널 스케일(kernel scale)이고, b는 바이어스 상수이다.where X is the input vector to be classified,

is the mean,

is the standard deviation,

is a linear predictor,

is a kernel scale, and b is a bias constant.

That is, the SVM binary classification unit 130 is a configuration that proceeds with classification using SVM after feature extraction in the feature extraction unit 120, and defines a hyper-plane dividing the two classes as shown in FIG. Binary classification is performed by maximizing the margin between the two classes based on the corresponding hyperplane.

The SVM learning is a process of finding a support vector capable of maximizing the margin.

10 is a flowchart illustrating a human body detection method using the human body detection radar system shown in FIG. 1 .

Referring to FIG. 10, in the human body detection method (S700) using the human body detection radar system according to an embodiment of the present invention, the signal processing unit 110 receives a frequency (radar) signal measured by a CW radar and STFT ( Signal processing step of generating a spectrogram through Short-Time Fourier Transform signal processing (S710), feature extraction step of extracting four features related to the movement of the human body based on the spectrogram in the feature extraction unit 120 (S720) ) and an SVM binary classification step (S730) of classifying the object as 1 or 0 using the above four features in the SVM binary classification unit.

Here, the frequency signal is a Doppler frequency signal that is a difference between a transmission frequency and a reception frequency for the target.

The signal processing step (S710) includes calculating the speed of the target using the Doppler frequency as shown in Equation 1 below.

[Equation 1]

V is the speed at which the object approaches the radar, and c is the speed of the radio wave.

In the signal processing step (S710), the radar signal is divided into short-time signals by applying a window function, and the Doppler frequency according to the time change is calculated by applying the FFT algorithm to the divided short-time signals, and the calculated result value It may be a step of generating the spectrogram by arranging on the time axis.

Next, the feature extraction step (S720) may be a step of extracting a plurality of features based on the time interval of micro-Doppler frequencies in the spectrogram.

For example, the period of human leg movement is different from that of short-legged dogs. The time it takes from moving one foot to moving the next is longer in long-legged people than in short-legged dogs. This can be confirmed by the micro-Doppler time interval on the spectrogram (see FIG. 5). That is, a plurality of features of the object may be extracted as an average value of time intervals of micro-Dopplers displayed vertically in FIG. 5 .

In addition, the feature extraction step (S720) may be a step of extracting a plurality of features based on whether micro-Doppler has occurred in the spectrogram.

More precisely, a plurality of features of the object may be extracted based on the number of strides of the object.

For example, as shown in FIG. 6 , unlike a human or a dog, a legless robot cleaner does not show micro Doppler caused by leg movements. This is suitable as a feature that distinguishes a person from an object without legs.

In addition, the feature extraction step (S720) may be a step of extracting a plurality of features using a radar reflection area within the spectrogram.

For example, radar cross section (RCS) is the total amount of energy that electromagnetic waves are reflected from an object. The reflected area varies depending on the size of the target, and the larger the target, the larger it appears. Therefore, the radar cross section (RCS) appears large in the order of people, dogs, and robot cleaners. Accordingly, a plurality of features of the object may be extracted through a value calculated by adding all data equal to or greater than a threshold value in the spectrogram.

In addition, the feature extraction step (S720) may be a step of extracting a plurality of features using a difference between a peak frequency and a body frequency in the spectrogram, and the peak frequency is the maximum of a micro-Doppler frequency. is the value

Here, the peak frequency is the maximum value of the micro-Doppler frequency and means the swing speed of the leg. The difference between the speed of the subject's torso and the swing speed of the legs is determined by the length of the subject's legs. Dogs have shorter legs than humans, so the difference between the body speed and the swing speed of the legs is smaller (see Fig. 7).

Accordingly, a plurality of features of the object may be extracted by calculating a difference between a trunk frequency indicating a trunk speed and a peak frequency indicating a leg swing speed on the spectrogram.

Next, the SVM binary classification step (S730) is a step of binary classification of four features through pre-learned SVM parameters.

In the SVM binary classification step (S730), features are classified into two classes using Equation 2 below.

[Equation 2]

는 평균,

는 표준편차,

는 선형 예측 변수,

는 커널 스케일(kernel scale)이고, b는 바이어스이다.where X is the input vector to be classified,

is the mean,

is the standard deviation,

is a linear predictor,

is the kernel scale, and b is the bias.

That is, the SVM binary classification step (S730) is a configuration in which the object is binary classified using the pre-learned SVM parameters after feature extraction in step S720, and as shown in FIG. 9, the hyper-plane separating the two classes ) and performing binary classification by maximizing the margin between the two classes based on the corresponding hyperplane, where the SVM learning is a process of finding a support vector capable of maximizing the margin.

Figure 1b is a block diagram showing a human body detection radar system for indoor security applications according to another embodiment of the present invention.

Referring to FIG. 1B , a human body detection radar system 200 according to another embodiment of the present invention includes a CW radar 10 and a signal processing system 210.

The CW radar 10 is a component that transmits a radar signal measuring an object.

The signal processing system 210 receives the radar signal and generates a spectrogram through STFT (Short-Time Fourier Transform) signal processing, the time interval of the fine Doppler frequency in the generated spectrogram, occurrence, radar reflection area, This is a configuration for classifying whether the target is a human as 1 or 0 using the difference between the peak frequency and the body frequency.

The frequency signal is a Doppler frequency signal that is a difference between a transmission frequency and a reception frequency for the target.

The signal processing system 210 includes a signal processing unit that calculates the speed of the target using the Doppler frequency, as shown in Equation 1 below.

[Equation 1]

V is the speed at which the object approaches the radar, and c is the speed of the radio wave.

The signal processing unit 110 divides the radar signal into short-time signals by applying a window function, calculates the Doppler frequency according to the time change for the divided short-time signals by applying the FFT algorithm, and then converts the calculated result into time signals. The spectrogram is created by arranging the axes.

The signal processing unit 110 extracts a plurality of features using any one of the time interval of the fine Doppler frequency in the generated spectrogram, whether or not it occurs, the radar reflection area, the difference between the peak frequency and the body frequency, and the variance value of the peak frequency. It includes a feature extraction unit 120 to.

The signal processing system 210 defines a hyperplane composed of two classes through learning and uses an SVM binary classification unit 130 that binary classifies the plurality of features by maximizing a margin between the two plane regions. include

The SVM binary classification unit 130 may be configured to maximize the margin by using a support vector closest to the hyperplane among feature vectors extracted by the feature extraction unit.

The margin (margine) means twice the distance of the distance from the chomyeongpyeon to the vector (support vector).

The SVM binary classification unit 130 classifies features into the two classes using Equation 2 below.

[Equation 2]

는 평균,

는 표준편차,

는 선형 예측 변수,

는 커널 스케일(kernel scale)이고, b는 바이어스이다.where X is the input vector to be classified,

is the mean,

is the standard deviation,

is a linear predictor,

is the kernel scale, and b is the bias.

Since each of the components 110, 120, and 130 of the signal processing system according to another embodiment of the present invention is substantially the same as the components of the embodiment shown in FIG. 1 (a), description of each component It is replaced with the description of FIG. 1 (a).

11 is a flowchart illustrating a human body detection method using a human body detection radar system for indoor security application according to another embodiment of the present invention.

Referring to FIG. 11, the human body detection method (S800) according to another embodiment of the present invention includes the steps of transmitting (S810) a radar signal measuring an object in a CW radar and receiving the radar signal to STFT (Short-Time Fourier Transform), a spectrogram is generated through signal processing, and the target is set to 1 or 0 using four factors: the time interval of fine Doppler frequencies in the generated spectrogram, whether or not it occurs, the radar reflection area, and the difference between the peak frequency and the body frequency. and a signal processing step of classifying (S820).

The frequency signal is a Doppler frequency signal that is a difference between a transmission frequency and a reception frequency for the object, and the signal processing step (S820) includes calculating the speed of the object using the Doppler frequency as shown in Equation 1 below. include

[Equation 1]

V is the speed at which the target approaches the radar, and c is the speed of the radio wave.

In the signal processing step (S820), the radar signal is divided into short-time signals by applying a window function, the Doppler frequency according to the time change is calculated for the divided short-time signals by applying the FFT algorithm, and the calculated result is and generating the spectrogram by arranging it on a time axis.

The signal processing step (S820) may include defining a hyperplane composed of two classes through SVM learning, and binary-classifying the plurality of features by maximizing a margin between the two plane regions.

The binary classification may be a step of classifying features into the two classes using Equation 2 below.

[Equation 2]

는 평균,

는 표준편차,

는 선형 예측 변수,

는 커널 스케일(kernel scale)이고, b는 바이어스이다.where X is the input vector to be classified,

is the mean,

is the standard deviation,

is a linear predictor,

is the kernel scale, and b is the bias.

Since step S820 of the human body detection method according to another embodiment of the present invention includes detailed steps in FIG. 10 , a more detailed description of S820 is replaced with a description of each step shown in FIG. 10 .

Hereinafter, with reference to FIGS. 12 to 14 , the contents of experiments for evaluating the classification performance of the system proposed in the present invention will be briefly described.

In order to confirm data acquisition and classification performance for SVM learning of the proposed system, experiments were conducted in indoor and outdoor environments as shown in FIG. 12 .

A total of 444 pieces of human, dog, and robot vacuum cleaner data of 235, 54, and 155 pieces were secured, and 215, 44, and 140 pieces were used as training sets for learning, respectively, and 20, 10, and 140 pieces were used as test sets for verification, respectively. 15 were used.

12 (a) is a photograph showing an outdoor human experiment, (b) an outdoor dog experiment, and (c) an indoor robot cleaner experiment.

13 shows a comparison of extracted feature values for each test subject, and a total of 120 pieces of data, 40 each for humans, dogs, and robot cleaners, were used. The characteristics of each of the 40 data were calculated and graphed.

13 (a) is a graph in which the first feature is calculated, and it can be seen that the highest value is obtained for a human and a value of 0 is obtained for a robot cleaner.

13(b) is a graph in which the second feature is calculated, and the number of steps of a person is counted, but in the case of a dog, the steps are too fast and are not properly counted.

13(c) and 13(d) are the third and fourth features, respectively, and are suitable as classification features because people and other objects are clearly distinguished.

13(e) is a graph showing the fifth feature, and it is clearly distinguished from a robot cleaner, but is not clearly distinguished from a dog and a human.

To determine which of the four features to use for classification, an experiment was conducted to check the accuracy.

14 is a table showing the average accuracy for each number of features used for classification, and FIG. 15 is a table showing the accuracy of the classification algorithm.

14 and 15, when all 5 features were used, the accuracy was 97.2%, when 4 features were used, 97.02%, when 3 features were used, 95.97%, and when 2 features were used, 93.54% showed accuracy. The more features you use, the higher the accuracy but the higher the computational complexity. For hardware implementation, considering the computational complexity and classification accuracy, it was designed to have good performance with low computational complexity by using only four features except for the fifth feature with the largest computational complexity.

Hereinafter, the hardware structure of the system proposed in the present invention will be described.

Referring to FIG. 2, the human body detection radar system 100, 200 of the present invention is largely composed of three blocks: STFT, Feature Extractor, and SVM Classifier. STFT consists of data memory, FFT block of Single Butterfly structure, Hanning Window ROM that stores window function, and DC Remover to remove DC component. It reads data from memory, removes DC component, and multiplies the window function to divide data. .

After 7 stages of FFT operation on the segmented data, the output stage delivers the FFT results to Feature Extractor. The calculated FFT results are used in parallel for four feature calculations in Feature Extractor. When the feature calculation is completed up to the last segmented data, the SVM Classifier block performs calculations for classification, and outputs 1 or 0 as a classification result to classify whether it is human or not.

Since only real-valued data comes out from the radar used, the FFT results come out symmetrically. Therefore, since the spectrogram also appears vertically symmetrically, only the upper half was used for feature point calculation to reduce memory usage. In addition, the spectrogram obtained as a result of STFT is not stored in memory, but reused in the feature extraction unit using a pipeline structure, so that it is implemented without using additional memory.

The proposed system was designed using Verilog HDL and verified and implemented using Altera Cyclone V 5CGXFC7C7F23C8 FPGA. It was implemented using only 1140 logics, 6.5Kb of memory, and 15 DSP blocks, and as a result of synthesis, it was confirmed that the total number of gates was 66K. 16 is a table showing a summary of resource use, and FIG. 17 is a table showing resource use by block. In addition, it was confirmed that it can operate with a maximum operating frequency of 104.79MHz, and the latency of the proposed design is 33,952 cycles, enabling real-time classification in a short time of about 0.324 ms to classify an object.

Therefore, the human body detection radar system for indoor security applications according to one embodiment and another embodiment of the present invention applies a signal processing using STFT, a feature extraction technique for efficient classification, and a classification algorithm using SVM, It simplifies complexity to minimize computational complexity, while providing the advantage of increasing the accuracy of human body detection.

That is, the human body detection radar system of the present invention extracts the spectrogram including time information by applying STFT from the CW radar and classifies it into SVM, so that the human body can be detected with high accuracy even with a small area.

18 is a diagram illustrating an example computing environment in which one or more embodiments set forth herein may be implemented, an illustration of a system 1000 that includes a computing device 1100 configured to implement one or more embodiments described above. shows For example, computing device 1100 may be a personal computer, server computer, handheld or laptop device, mobile device (mobile phone, personal digital assistant, media player, etc.), multiprocessor system, consumer electronics, mini computer, mainframe computer, distributed computing environments that include any of the foregoing systems or devices; and the like.

Computing device 1100 may include at least one processing unit 1110 and memory 1120 . Here, the processing unit 1110 may include, for example, a central processing unit (CPU), a graphics processing unit (GPU), a microprocessor, an Application Specific Integrated Circuit (ASIC), a Field Programmable Gate Arrays (FPGA), and the like. and may have a plurality of cores. The memory 1120 may be volatile memory (eg, RAM, etc.), non-volatile memory (eg, ROM, flash memory, etc.), or a combination thereof. Additionally, computing device 1100 may include additional storage 1130 . Storage 1130 includes, but is not limited to, magnetic storage, optical storage, and the like. The storage 1130 may store computer readable instructions for implementing one or more embodiments disclosed herein, and may also store other computer readable instructions for implementing an operating system, application programs, and the like. Computer readable instructions stored in storage 1130 may be loaded into memory 1120 for execution by processing unit 1110 . Computing device 1100 can also include input device(s) 1140 and output device(s) 1150 .

Here, input device(s) 1140 may include, for example, a keyboard, mouse, pen, voice input device, touch input device, infrared camera, video input device, or any other input device. Output device(s) 1150 may also include, for example, one or more displays, speakers, printers, or any other output devices, or the like. Additionally, computing device 1100 may use an input device or output device included in another computing device as input device(s) 1140 or output device(s) 1150 .

Computing device 1100 may also include communication connection(s) 1160 that allow computing device 1100 to communicate with other devices (eg, computing device 1300).

Here, communication connection(s) 1160 may be a modem, network interface card (NIC), integrated network interface, radio frequency transmitter/receiver, infrared port, USB connection, or other device for connecting computing device 1100 to other computing devices. May contain interfaces. Further, communication connection(s) 1160 may include a wired connection or a wireless connection. Each component of the aforementioned computing device 1100 may be connected by various interconnections such as a bus (eg, peripheral component interconnection (PCI), USB, firmware (IEEE 1394), optical bus structure, etc.) and may be interconnected by the network 1200. Terms such as "component" and "system" as used herein generally refer to a computer-related entity that is hardware, a combination of hardware and software, software, or software in execution.

For example, a component may be, but is not limited to, a process running on a processor, a processor, an object, an executable, a thread of execution, a program, and/or a computer. For example, both the application running on the controller and the controller may be components. One or more components can reside within a process and/or thread of execution and a component can be localized on one computer or distributed between two or more computers.

In the above, the present invention has been described in detail with reference to the examples, but those skilled in the art to which the present invention belongs will be able to make various substitutions, additions, and modifications without departing from the technical idea described above. Of course, it should be understood that such modified embodiments also belong to the protection scope of the present invention, which is defined by the appended claims below.

100, 200: human body detection radar system
110: signal processing unit
120: feature extraction unit
130: SVM binary classification unit

Claims (33)

A signal processing unit for receiving a radar signal measured by a CW radar and generating a spectrogram through short-time Fourier transform (STFT) signal processing;
a feature extractor extracting four features based on the spectrogram; and
An SVM binary classification unit for classifying whether the target is a human as 1 or 0 using all of the four features,
The feature extraction unit
Characteristics of the time interval of the micro-Doppler frequency in the spectrogram, the count of stride of the target, the reflection area of the CW radar, and the difference between the peak frequency and the body frequency in the spectrogram Extraction, the peak frequency is the maximum value of the micro-Doppler frequency,
The SVM binary classification unit
A human body detection radar system that performs binary classification through the following equation using a parameter learned to maximize a margin.
[ceremony]

(Where X is the input vector to be classified,

is the mean,

is the standard deviation,

is a linear predictor,

is the kernel scale, and b is the bias.)
According to claim 1,
The frequency signal is a Doppler frequency signal that is a difference between a transmission frequency and a reception frequency for the target,
The signal processing unit
A human body detection radar system that calculates the speed of the object using the Doppler frequency as in the following equation.
[ceremony]

(V is the speed at which the object approaches the radar, c is the speed of the radio wave.)
According to claim 1,
The signal processing unit
The radar signal is divided into short-time signals by applying a window function, the Doppler frequency according to the time change is calculated for the divided short-time signals by applying the FFT algorithm, and the calculated results are arranged on the time axis to form the spectrogram Human body detection radar system, characterized in that for generating.
delete delete delete delete delete According to claim 1,
The margin is
A human body detection radar system that means twice the distance from the hyperplane to the vector (support vector).
A signal processing step of generating a spectrogram through STFT (Short-Time Fourier Transform) signal processing by receiving a frequency (radar) signal measured by a CW radar in a signal processing unit;
a feature extraction step of extracting four features based on the spectrogram in a feature extraction unit; and
An SVM binary classification step of classifying whether the object is a human as 1 or 0 using all the four features in an SVM binary classification unit,
The feature extraction step is
Extracting a plurality of features based on a time interval of a micro-Doppler frequency in the spectrogram, a count of stride of the target, a radar reflection area, and a difference between a peak frequency and a body frequency ego,
The peak frequency is the maximum value of the micro-Doppler frequency,
The SVM binary classification step is
A method of operating a human body detection radar system, characterized in that the step of classifying the learned parameters by maximizing a margin through the following equation.
[ceremony]

(Where X is the input vector to be classified,

is the mean,

is the standard deviation,

is a linear predictor,

is the kernel scale, and b is the bias.)
According to claim 10,
The frequency signal is a Doppler frequency signal that is a difference between a transmission frequency and a reception frequency for the target,
The signal processing step is
As shown in the following equation, a method of operating a human body detection radar system that calculates the speed of the target using the Doppler frequency.
[ceremony]

(V is the speed at which the object approaches the radar, c is the speed of the radio wave.)
According to claim 10,
The signal processing step is
The radar signal is divided into short-time signals by applying a window function, the Doppler frequency according to the time change is calculated for the divided short-time signals by applying the FFT algorithm, and the calculated results are arranged on the time axis to form the spectrogram Method of operating a human body detection radar system, characterized in that the step of generating.
delete delete delete delete delete According to claim 10,
The margin is
A method of operating a human body detection radar system, which means twice the distance from the hyperplane to the vector (support vector).

delete delete delete delete delete delete delete delete delete delete delete delete delete delete delete

Images