CN105629228A - Partition human body motion detection method based on k-means clustering and Bayes classification - Google Patents

Abstract

The invention discloses a partition wall human motion detection method based on K-means clustering and Bayesian classification. The method first receives the precoded signal waveform from the transmitter through the receiver; secondly, the received signal is divided into a section Segment signal, perform short-time Fourier transform on each segment signal to obtain the transformation matrix, and calculate the variance vector of the transformation matrix to obtain the extreme difference value; finally put the extreme difference value into the pre-clustered training data and use Bayesian Classification is carried out in the Bayesian classifier obtained by classification, and whether the human body in the partition wall is moving is judged according to the classification result. The invention adopts K-means clustering and Bayesian classification, can effectively detect whether the human body moves in the partition wall, and greatly improves the accuracy of detecting the human body motion in the partition wall.

Description

Human motion detection method based on K-means clustering and Bayesian classification

technical field

The invention relates to a method for detecting human motion on a partition wall, more specifically to a method for detecting human motion on a partition wall based on K-means clustering and Bayesian classification.

technical background

Human body detection within the general line of sight can be detected using optoelectronic devices such as infrared cameras and cameras. These techniques are commonly used in intrusion detection in art galleries and banks. However, these technologies have great limitations and cannot be used to detect objects behind non-transparent walls (or shelters) such as stone, wood, concrete, etc., so the detection technology used must have a perspective effect. At present, detection technologies with perspective effects are commonly based on X-rays and ultrasonic echoes, but none of these perspective technologies can well meet the current needs for human body detection through walls. X-rays are high-energy rays. Although they can penetrate walls, they can cause great damage to the human body; while ultrasonic echoes have relatively large attenuation to layered media. To sum up, it is very feasible to use specific frequency electromagnetic waves with good penetration to the wall and negligible harm to the human body as the emission signal for the detection of human motion on the partition wall. As a transmitting signal, electromagnetic waves can penetrate non-metallic media such as wooden doors and concrete walls, and realize the detection of moving targets behind the walls.

In special operations such as anti-riot and emergency rescue, whether the human body movement information in the room or behind the wall can be effectively detected will have a major impact on combat and rescue, and can greatly reduce the number of casualties. Therefore, detection technology capable of detecting objects behind non-metallic and transparent media such as walls and wooden doors has received more and more attention.

Although the traditional wall-penetrating ultra-wideband radar can detect human movement in the partition wall, it occupies a large amount of bandwidth, has a large transmission power, and has a very large antenna array. However, it is very challenging to use wireless communication devices with small bandwidth, low transmission power, and small size to detect human motion across walls. It is necessary to detect weak targets under strong noise. At present, the technology of the partition wall human motion detection method realized by this portable device needs to be further studied and discussed.

Contents of the invention

The object of the present invention is to propose a partition wall human motion detection method based on K-means clustering and Bayesian classification, which can effectively detect whether the partition wall human body moves and improve detection accuracy.

The object of the present invention is achieved by the following technical solutions: a partition wall human motion detection method based on K-means clustering and Bayesian classification, the method may further comprise the steps:

Step 1, arrange the first transmitter, second transmitter and receiver on one side of the wall; first the first transmitter sends the original signal, after the receiver receives the signal, the second transmitter sends the same original signal, and the receiver receives signal; then calculate the precoded signal of the second transmitter through the two received signals; the last two transmitters transmit signals at the same time, the first transmitter sends the original signal, and the second transmitter sends the precoded signal;

Step 2, the receiver receives the superimposed signals sent by the two transmitters at the same time, and evenly divides the received signals according to time;

Step 3, perform short-time Fourier transform on each segment of the signal segmented in step 2 to obtain a short-time Fourier transform matrix A _m×n , m represents the number of frequency points of Fourier transform (FFT), and n is The number of time points of each segment of signal calculated according to the size of the window function and the number of overlaps, the element A _ij in the matrix represents the short-time Fourier transform value at i frequency and j time point;

Step 4, perform variance statistics on the short-time Fourier transform matrix A _m×n obtained in step 3, that is, calculate the variance v _j of the short-time Fourier transform values corresponding to all frequency points at each time point, and finally get this The variance vector v _1×n at all time points of the segment signal; at the same time, perform absolute median difference statistics on the short-time Fourier transform matrix A _m×n to obtain the absolute median difference vector MAD _1×n ;

Step 5, calculate the range value v _range of the variance vector v _1×n , that is, v _range =v _max -v _min , where v _max is the maximum value in the variance vector v _1×n , and v _min is the variance vector v _1×n The minimum value in; similarly calculate the absolute median difference vector MAD _1×n range value MAD _range ;

Step 6, according to steps 1-5, calculate the extreme difference v' _range and MAD' _range when the partition wall is moved by people and the range value v″ _range and MAD" _range when no one is moving on the partition wall; use K mean value (Kmeans ) method clusters and aggregates the range values in two cases into two clusters, and uses the range values and clustering results as a training set for Bayesian classification to obtain a Bayesian classifier;

Step 7, when detecting human motion on the partition wall, calculate the range value v _range and MAD _range of a section of signal according to steps 1-5, and put the range value v _range and MAD _range into the Bayesian classifier obtained in step 6 Classify, if the Bayesian classifier classifies it as the human body movement of the partition wall, then the human body is moving at the partition wall at this moment; and if it is classified into another category, there is no human body moving on the partition wall; This step is repeated for each segment of the signal, so that the moment of human motion on the partition wall can be given.

The method for detecting human body movement on a partition wall based on K-means clustering and Bayesian classification in the present invention can detect whether there is human body movement on the partition wall. Compared with the prior art, the present invention has the following advantages:

1. Use K-means clustering and Bayesian classification to calculate the similarity of signal waveforms. Compared with threshold detection and other detection methods, it is not necessary to select an appropriate threshold; instead, the classifier is used for classification, and the classification standard is determined by the classification device decision;

2. It can realize real-time detection, carry out corresponding signal processing according to the received signal, and give the result of the detected motion direction in real time;

3. It can adapt to different environments and different human body movement patterns without making corresponding changes in advance for changes in the environment and movement patterns;

4. The detection blind area is small, and the detection can be realized in the effective detection area.

Description of drawings

Fig. 1 is the flowchart of transmitter and receiver;

Fig. 2 is the flow chart of processing the partition wall human motion detection signal based on K-means clustering and Bayesian classification;

Figure 3 is the K-means clustering results.

detailed description

The present invention will be described in further detail below in conjunction with the accompanying drawings.

The present invention provides a partition wall human motion detection method based on K-means clustering and Bayesian classification. The process of sending and receiving signals is shown in Figure 1. Two transmitters and one receiver are used. . First, the first transmitter sends a signal, and the receiver receives the signal; secondly, the second transmitter sends the same signal as the first transmitter, and the receiver receives the signal; then, according to the two received signals, calculate the precoding The signal; Finally, let the two transmitters send the signal at the same time, and the receiver receives the signal. Here, the first transmitter still sends the original signal, while the second transmitter sends the precoded signal just calculated.

On the basis of above-mentioned signal sending and receiving, detection method of the present invention, as shown in Figure 2, comprises the following steps:

Step 1, first let the receiver and two transmitters run on one side of the wall for a period of time, the receiver will receive signals superimposed by various reflection signals from behind the wall and on this side of the wall;

Step 2, evenly divide the received signal according to time, and divide it into small signals of segments, here, it is specifically divided into 1s signal data;

Step 3, perform short-time Fourier transform (STFT) STFT(t,ω)=∫s(t')ω(t'-t)e ^-jωt' dt' on each segment of the divided small signal to obtain a The short-time Fourier transform matrix A _m×n , the number of rows of the matrix m represents the number of points of the Fourier transform (FFT), that is, how many frequency points; and the number of columns of the matrix n is based on the window function The number of time points of each small signal obtained by calculating the size and the number of overlaps. Therefore, the transformation matrix is not only related to frequency, but also related to time. The element A _ij in the matrix represents the short-time Fourier transform value at i frequency and j time point;

Step 4, perform variance statistics on the short-time Fourier transform matrix A _m×n obtained in step 3, that is, calculate the variance v _j of the short-time Fourier transform values corresponding to all frequency points at each time point, and finally get this The variance vector v _1×n at all time points of the segment signal; at the same time, carry out absolute median difference statistics on the transformation matrix A _m×n to obtain the absolute median difference vector MAD _1×n ;

Step 5, calculate the range value v _range of the variance vector v _1×n , that is, v _range =v _max -v _min , where v _max is the maximum value in the variance vector v _1×n , and v _min is the variance vector v _1×n The minimum value in; similarly calculate the absolute median difference vector MAD _1×n range value MAD _range ;

Step 6, according to steps 1-5, calculate the extreme difference v' _range and MAD' _range when the partition wall is moved by people and the range value v″ _range and MAD" _range when no one is moving on the partition wall; use K mean value (Kmeans ) method clusters and aggregates the range values in two cases into two clusters, and uses the range values and clustering results as a training set for Bayesian classification to obtain a Bayesian classifier;

Step 7. When detecting human motion on the partition wall, calculate the range value v _range and MAD _range of a section of signal according to steps 1-5, and put the range value v _range and MAD _range into the Bayesian classifier obtained in step 6 Classify, if the Bayesian classifier classifies it as the human body movement of the partition wall, then the human body is moving at the partition wall at this moment; and if it is classified into another category, there is no human body moving on the partition wall; This step is repeated for each segment of the signal, so that the moment of human movement on the partition wall can be given.

The technical scheme adopted by the detection method of the present invention is: carry out K-means clustering to the training data into two clusters in advance, then obtain a Bayesian classifier by Bayesian classification of the training data and clustering results, and then use the received The signal is divided into sections of signals, the short-time Fourier transform is performed on each section of the signal to obtain the transformation matrix, and the variance vector of the transformation matrix is calculated to obtain the range value; finally, the range value is put into the Bayesian classifier to obtain the decision classification results.

The invention adopts a transmitter with small bandwidth and low transmission power to realize the movement of the human body at the partition wall and ensure the detection accuracy. Compared with the traditional wall-penetrating ultra-wideband radar which occupies a large amount of bandwidth, high transmission power and very large antenna array, the present invention has significant advantages.

Example

Arrange two transmitters and one receiver on one side of the wall, and the moving body walks randomly on the other side of the wall. The two transmitters and receivers are arranged equidistantly on the same horizontal plane and at the same distance from the wall. The wall body of the experiment is a 25cm thick concrete wall, and its attenuation is 20dB. The bandwidth of the transmitter is 1MHz, the transmission power is 100mW, the transmission frequency is 2.4GHz, and it contains 3 directional antennas. In order to make the movement pattern simpler and more regular, two movement patterns are defined, 1) walking parallel to the wall and 2) walking perpendicular to the wall. The K-means clustering results are shown in Figure 3.

According to the method of the present invention, the detection rate of human body movement on the partition wall can reach 90%. Compared with the traditional wall-penetrating ultra-wideband radar that occupies a large amount of bandwidth and high transmission power, the method of the present invention also has the advantages of narrow bandwidth and low transmission power. High detection accuracy.

Claims (1)

1. the partition wall body movement detection method based on K mean cluster and Bayes's classification, it is characterised in that the method comprises the following steps:

Step 1, arranges the first transmitter, the second transmitter and receiver in the side of wall; First the first transmitter sends primary signal, and after receiver receives signal, the second transmitter sends same primary signal, and receiver receives signal; The signal then passing through twice reception calculates the precoded signal of the second transmitter; Signal launched by last two transmitters simultaneously, and the first transmitter sends primary signal, and the second transmitter sends precoded signal;

Step 2, receiver receives the signal after the superposition that two transmitters send simultaneously, and temporally carries out even partition to the received signal;

Step 3, carries out Short Time Fourier Transform to every segment signal of step 2 segmentation, obtains a Short Time Fourier Transform matrix A_m��n, m represents the Frequency point number of Fourier transformation (FFT), and n is the time point number according to window function size and the overlapping calculated every segment signal of number, the elements A in matrix_ijRepresent in i frequency, the Short Time Fourier Transform value of j time point;

Step 4, the Short Time Fourier Transform matrix A that step 3 is obtained_m��nCarry out variance statistic, namely calculate the variance v of Short Time Fourier Transform value corresponding to all Frequency points on each time point_j, finally give the variance vectors v on all time points of this segment signal_1��n; Simultaneously to Short Time Fourier Transform matrix A_m��nCarry out median absolute deviation statistics, obtain median absolute deviation vector M AD_1��n;

Step 5, calculates variance vectors v_1��nExtreme difference value v_range, i.e. v_range=v_max-v_min, v_maxFor variance vectors v_1��nIn maximum, v_minFor variance vectors v_1��nIn minima; In like manner calculate median absolute deviation vector M AD_1��nExtreme difference value MAD_range;

Step 6, calculates the extreme difference value v ' when partition wall has people to move according to step 1-5 respectively_range��MAD'_rangeExtreme difference value v during motion unmanned with partition wall "_range��MAD"_range; Adopt K average (Kmeans) method that the extreme difference value in two kinds of situations is carried out cluster and aggregate into two bunches, and extreme difference value and cluster result are carried out Bayes's classification as training set, obtain a Bayes classifier;

Step 7, when carrying out partition wall human motion detection, calculates the extreme difference value v of a segment signal according to step 1-5_rangeAnd MAD_range, by extreme difference value v_rangeAnd MAD_rangePutting into the Bayes classifier that step 6 obtains to classify, if Bayes classifier is categorized into partition wall human motion situation, then this moment partition wall human body is in motion; And being classified as another kind of, then partition wall does not have human body in motion; Every segment signal of step 2 segmentation is repeated this step, such that it is able to provide the moment of partition wall human motion.