A method for comprehensively distinguishing the types of living things in a cabin based on multiple features

Abstract

The present invention belongs to the field of automobile radar technology, and specifically relates to a method for comprehensively distinguishing the types of living things in a cabin based on multiple features. By improving the signal processing algorithm without increasing hardware resources, the FOV detection area is improved through DBF, and the detection area of ​​a single radar can cover the front and rear rows of seats in a passenger car, effectively reducing the hardware cost and wiring complexity of the car factory. The type of living things in the cabin is comprehensively distinguished by multi-dimensional features, which can significantly improve the detection accuracy and anti-interference ability of personnel types in complex scenes. Through actual vehicle testing, the comprehensive accuracy can reach more than 95%. The radar is small in size, light in weight, and has a simple interface. It can be used in all scenarios, is not affected by the size of the sunroof in the car, and can be adapted to various models of the car factory.

Description

Technical Field

[0001] The present invention belongs to the field of automotive radar technology, and specifically relates to a method for comprehensively distinguishing the type of living things in a cabin based on multiple features. Background Technology

[0002] With the popularity of passenger cars in China, the current number of passenger cars in my country has exceeded 400 million. Millimeter-wave radar, as an active non-contact sensor, can sense the distance, angle and speed information of surrounding objects. Its distance perception resolution based on the phase method can theoretically reach the sub-millimeter level, and can accurately measure the human chest breathing and limb micro-movement signals. In addition, due to its non-imaging perception characteristics, millimeter-wave radar has a good privacy protection function. Therefore, it is an ideal sensor for sensing life in the automotive cabin environment (also known as "cabin liveness detection radar"), especially suitable for scenarios where children are detected in the cabin. Secondly, since the European Union and the China New Car Assessment and Evaluation Association (ENCAP and CNCAP) both emphasize the detection and alarm capabilities when only children are in the cabin, improving the recognition accuracy of the type of living things in the cabin is the key to improving the success rate of detection and alarm for the presence of children in the cabin.

[0003] In the prior art, the general working principle and process of life presence detection are as follows: Assuming that the current vehicle is a double-row five-seater vehicle, the cabin liveness detection radar is installed in the middle of the cabin roof, and signals are transmitted and received at a certain frame repetition frequency, with the azimuth dimension corresponding to the left and right direction of the vehicle, and the pitch dimension corresponding to the front and rear direction of the vehicle; the target echo signal obtained by each receiving channel is processed by Fast Fourier Transform (Fast Fourier Transform, FFT) in the distance dimension and Doppler dimension, and the zero Doppler signal is extracted; several frames of zero Doppler signals are accumulated, and a threshold is set to obtain a range gate position with a higher amplitude; the range gate position is traversed, the multi-frame zero Doppler signal is phase unwrapped, and the corresponding breathing frequency is calculated by FFT for life detection; the detected breathing frequency is compared with the breathing frequency-age comparison table to obtain the type of person.

[0004] From the above, it can be seen that the life presence detection method has the following defects: the detection method requires the tested person to be in a static state, and when the person moves, it will interfere with the detection result; the individual differences in breathing rate are large, and the breathing rate of children (12-16 years old) is easily mixed with that of adults; in a multi-target scenario, it is difficult to detect the breathing rate of all people. SUMMARY OF THE INVENTION

[0005] The purpose of the present invention is to provide a method for comprehensively distinguishing the type of living things in a cabin based on multiple features, and to detect the type of living things in the cabin based on a multi-transmit and multi-receive millimeter-wave radar in a time-division multiplexing mode, so as to solve the problems raised in the above-mentioned background technology.

[0006] To achieve the above object, the present invention adopts the following technical solutions:

[0007] The present invention proposes a method for comprehensively discriminating the types of in-cabin living bodies based on multiple features, including:

[0008] S1. Install a millimeter-wave radar on the roof inside the cabin, configure the signal transmission and reception of the millimeter-wave radar according to the antenna and waveform configuration, perform 2D-FFT processing on the received echo data, and extract zero-Doppler data;

[0009] S2. Select a set of azimuth and elevation channels respectively from the zero-Doppler data, and perform DBF processing on the zero-Doppler signals of several consecutive frames. Among them, the DBF angles are selected as the azimuth angles corresponding to the driver's and passenger's seat positions and the elevation angles corresponding to the front and rear row seat positions;

[0010] S3. Take out the peaks of the two-dimensional DBF accumulation results and multiply and add them correspondingly to obtain the amplitude results P front and P rear , and obtain P front and P rear when there are people in the front and rear rows based on the pre-collected data as empirical thresholds, and compare the measured results with the empirical thresholds. If the measured front row or result is greater than or , then add the front row or the rear row to the living body detection area;

[0011] S4. Perform additional DBF processing on the elevation zero-Doppler in S2, and the selected DBF angle is the elevation angle corresponding to the headrest position of the front row or the rear row;

[0012] S5. The processed zero-Doppler accumulation result in S3 is S DBF , where S DBF is a 1×N vector;

[0013] S6. Select the distance gate serial number (IDX m ) corresponding to the headrest position as the demarcation point, and search for the maximum (P DBF in S m and P 1 and P 2 ) and the cumulative peak values of the two partitions before and after IDX

[0014] S7. Based on the measured data, perform linear fitting on P 1 , P 2 , and , and obtain the fitting coefficients and corresponding weights (W 1 , W2 , W 3 ), construct a classifier, and at the same time combine other dimensional features and corresponding weights (W 4 ) to comprehensively discriminate the types of in-cabin living bodies;

[0015] S8. Output the type discrimination result with reference to the discrimination standard.

[0016] Preferably, the configuration of the antenna includes: building a hardware platform using a general millimeter-wave SOC chip, and setting a radar receiving channel in the hardware platform; the configuration of the waveform adopts the TDM-MIMO time division multiplexing method.

[0017] Preferably, the process of performing 2D-FFT processing on the received echo data and extracting zero-Doppler data includes: transmitting waves based on the configuration of the antenna and waveform; performing 2D-FFT processing on the echo signal by the hardware accelerator in the hardware platform; obtaining a signal matrix with a distance dimension length of m and a Doppler dimension length of n after hardware acceleration processing; obtaining zero-Doppler data.

[0018] Preferably, only the signals at the positions where the Doppler value is zero are retained in the signal matrix.

[0019] Preferably, the process of performing DBF processing on the zero-Doppler signals of several consecutive frames includes:

[0020] Selecting the zero-Doppler signals of the first group of azimuth dimension channels and the first group of elevation dimension channels in the receiving channel respectively for DBF processing, and the formula is as follows:

[0021]

[0022] Among them, P DBF (θ) is the DBF energy spectrum, a(θ) is the steering vector, R is the input signal covariance, L is the number of channels, and θ is the selected DBF angle.

[0023] Preferably, the discrimination standard includes:

[0024] A. When the in-cabin living body detection radar is powered on and working, based on the P 1 , P 2 , and features obtained from the actual measurement results in the above steps, input them into the classifier for type discrimination, and output the discrimination result and confidence level, where the confidence level is obtained by calculating the L2 norm of the actual measurement features and the fitting parameters;

[0025] B. When the classification result of the classifier conflicts with other dimensional features such as the breathing frequency, sort the classification results obtained from each feature based on the weights and confidence levels of different classification features, and output the classification result with the largest product of the two.

[0026] Technical effects and advantages of the present invention: A method for comprehensively discriminating the types of in-vehicle living bodies based on multiple features proposed by the present invention has the following advantages compared with the prior art:

[0027] Without adding hardware resources, the present invention improves the signal processing algorithm. By using DBF, the FOV detection area is enhanced. The detection area of a single radar can cover the front and rear row seats of a passenger car, effectively reducing the hardware cost and wiring complexity of the vehicle factory. By comprehensively discriminating the types of in-vehicle living bodies using multi-dimensional features, the detection accuracy and anti-interference ability of personnel types in complex scenarios can be significantly improved. Through real vehicle tests, the comprehensive accuracy rate can reach over 95%. Moreover, the radar is small in size, light in weight, has a simple interface, can be used in all scenarios, is not affected by the size of the sunroof in the vehicle, and can be adapted to various vehicle models of the vehicle factory. Description of the Drawings

[0028] Figure 1 It is a schematic structural diagram of the hardware platform of the present invention;

[0029] Figure 2 It is a distribution diagram of the receiving channels of the present invention;

[0030] Figure 3 It is a schematic diagram of waveform configuration of the present invention;

[0031] Figure 4 It is a schematic diagram of the selection of the pitch-dimensional DBF angle of the present invention;

[0032] Figure 5 It is a schematic diagram of the zero Doppler accumulation result of the present invention;

[0033] Figure 6 It is a flowchart of the method for comprehensively discriminating the types of in-vehicle living bodies based on multiple features of the present invention. Detailed Embodiments

[0034] Next, the technical solutions in the embodiments of the present invention will be clearly and completely described in conjunction with the accompanying drawings in the embodiments of the present invention. Obviously, the described embodiments are only a part of the embodiments of the present invention, rather than all of the embodiments. The specific embodiments described herein are only used to explain the present invention and are not used to limit the present invention. Based on the embodiments of the present invention, all other embodiments obtained by those of ordinary skill in the art without creative efforts shall fall within the protection scope of the present invention.

[0035] In the embodiments of the present invention, there is provided a method for comprehensively discriminating the types of in-vehicle living bodies as shown in Figure 6 and includes:

[0036] S1. Install the millimeter-wave radar on the roof inside the cabin, configure the signal transmission and reception of the millimeter-wave radar according to the antenna and waveform configuration, perform 2D-FFT processing on the received echo data, and extract zero-Doppler data;

[0037] Specifically, the configuration of the antenna includes: building a hardware platform using a general-purpose millimeter-wave SOC chip, and setting up radar receiving channels in the hardware platform; the configuration of the waveform adopts the TDM-MIMO time-division multiplexing method;

[0038] Among them, the hardware platform uses a general-purpose millimeter-wave SOC chip platform. Here, the AWR6843AOP chip (AWR6843) of Texas Instruments is taken as an example. The hardware resources and radar parameters are respectively as Figure 1 shown in Table 1, and the distribution of the radar receiving channels is as Figure 2 shown.

[0039] Table 1:

[0040]

[0041] The waveform configuration adopts the TDM-MIMO time-division multiplexing method (TX0→TX1→TX2), and the wave transmission settings for each pulse are as Figure 3 shown.

[0042] Furthermore, when performing 2D-FFT processing on the received echo data and extracting zero-Doppler data, install the radar at the center inside the vehicle cabin or at the center of the rear edge of the front row seat back. Based on Figure 1 the hardware platform and Figure 3 the waveform configuration for wave transmission, the hardware accelerator in the hardware platform performs 2D-FFT processing on the echo signal. After hardware acceleration processing, a signal matrix with a distance dimension length of m and a Doppler dimension length of n is obtained. To save hardware resources, only the signals (zero-Doppler data) where the Doppler value is zero are retained for subsequent processing in this method.

[0043] S2. Select a group of azimuth and elevation channels respectively from the zero-Doppler data, and perform DBF processing on the zero-Doppler signals of several consecutive frames. Among them, the DBF angle selections are respectively the azimuth angles corresponding to the driver and passenger seat positions and the elevation angles corresponding to the front and rear row seat positions;

[0044] Specifically, in the receiving channels of AWR6843, as Figure 2 shown, select the zero-Doppler signals of the first group of azimuth channels (sequence numbers 12, 10, 9, 6) and the first group of elevation channels (sequence numbers 4, 3, 12, 11) respectively for DBF processing. The formula is as follows:

[0045]

[0046] Among them, P DBF (θ) is the DBF energy spectrum, a(θ) is the steering vector, R is the covariance of the input signal, L is the number of channels, and θ is the selected DBF angle.

[0047] S3. Extract the peaks of the two-dimensional DBF cumulative results and multiply and add them correspondingly to obtain the corresponding amplitude results P front and P rear , and obtain P front and P rear when there are people in the front row and the back row based on the pre-collected data as the empirical thresholds, and compare the measured results with the empirical thresholds. If the measured front row or result is greater than or , then add the front row or the back row to the in vivo detection area.

[0048] S4. Perform additional DBF processing on the zero Doppler in the pitch dimension in S2, and select the pitch angle corresponding to the headrest position in the front row or the back row as the DBF angle, as shown in Figure 4 .

[0049] S5. Take the processed zero Doppler cumulative result in S3 as S DBF , where S DBF is a 1×N vector, as shown in Figure 5 .

[0050] S6. Select the distance gate number (IDX m ) corresponding to the headrest position as the demarcation point, and search for the maximum values (P DBF in IDX m before and after the two partitions in S 1 and P 2 ) and the cumulative peak value as the classification feature, as shown in Figure 5 .

[0051] S7. Based on the measured data, perform linear fitting on P 1 , P 2 , and , and obtain the fitting coefficients and the corresponding weights (W 1 , W 2 , W 3 ), construct a classifier, and comprehensively discriminate the in vivo types in the cabin by combining other dimensional features (such as breathing frequency, point cloud height, etc.) and the corresponding weights (W 4 );

[0052] S8. Output the type discrimination result with reference to the discrimination criteria. Specifically, the discrimination criteria include:

[0053] A. When the in-cabin living body detection radar is powered on and working, the measured P 1 , P 2 , and features are obtained. These features are input into a classifier for type discrimination, and the discrimination result and confidence level are output. The confidence level is obtained by calculating the L2 norm of the measured features and the fitting parameters;

[0054] B. When the classification result of the classifier conflicts with other dimension features such as the breathing frequency, the classification results obtained from each feature are sorted based on the weights and confidence levels of different classification features, and the classification result with the largest product of the two is output.

[0055] In this embodiment, without increasing hardware resources, the signal processing algorithm is improved, and the FOV detection area is enhanced through DBF. The detection area of a single radar can cover the front and rear row seat areas of a passenger car, effectively reducing the hardware cost and wiring complexity of the vehicle factory. By comprehensively discriminating the type of in-cabin living body using multi-dimensional features, the detection accuracy and anti-interference ability of the personnel type in complex scenarios can be significantly improved. Through actual vehicle tests, the comprehensive accuracy rate can reach more than 95%. Moreover, the radar is small in size, light in weight, has a simple interface, can be used in all scenarios, is not affected by the size of the sunroof in the vehicle, and can be adapted to various vehicle models of the vehicle factory.

[0056] Finally, it should be noted that the above are only the preferred embodiments of the present invention and are not used to limit the present invention. Although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art can still modify the technical solutions recorded in the foregoing embodiments, or perform equivalent replacements on some of the technical features. Any modifications, equivalent replacements, improvements, etc. made within the spirit and principle of the present invention shall be included in the protection scope of the present invention.

Claims

1. A method for comprehensively distinguishing the type of living things in a cabin based on multiple features, characterized in that: include: S1. Install the millimeter-wave radar on the roof of the cabin, and configure the signal transmission and reception of the millimeter-wave radar according to the antenna and waveform configuration, complete 2D-FFT processing on the received echo data, and extract zero Doppler data; S2. Selecting a set of azimuth and pitch channels from the zero Doppler data, and performing DBF processing on a number of consecutive frames of zero Doppler signals, wherein the DBF angles are the azimuth angles corresponding to the positions of the driver and co-driver seats and the pitch angles corresponding to the positions of the front and rear seats, respectively; S3. Take out the peak values ​​of the two-dimensional DBF accumulation results and multiply and add them accordingly to obtain the amplitude results corresponding to the front and back rows respectively. and , based on the pre-collected data, the and As the empirical threshold, the measured results and Compare with the empirical threshold, if the front row result is measured Greater than , then add the front row to the liveness detection area; if the measured back row result Greater than When , the back row is added to the living body detection area; S4, performing additional DBF processing on the pitch dimension zero Doppler in S2, wherein the DBF angle is selected as the pitch angle corresponding to the position of the front row or rear row headrest; S5, the processed zero Doppler accumulation result in S3 is ,in for A vector of S6. Select the distance gate number corresponding to the headrest position As the dividing point, search middle The maximum of the two partitions and and cumulative and Peak value, , , and As a classification feature; S7, based on the measured data , , and Perform linear fitting, obtain fitting coefficients and corresponding weights, and construct a classifier. At the same time, combine the respiratory rate and corresponding weights to comprehensively identify the type of living organisms in the cabin. S8. Output the type discrimination result by referring to the discrimination standard.

2. The method for comprehensively distinguishing the type of living things in a cabin based on multiple features according to claim 1 is characterized in that: The configuration of the antenna includes: using a general millimeter wave SOC chip to build a hardware platform, and setting a radar receiving channel in the hardware platform; the configuration of the waveform adopts a TDM-MIMO time division multiplexing method.

3. The method for comprehensively distinguishing the type of living things in a cabin based on multiple features according to claim 2 is characterized in that: The step of performing 2D-FFT processing on the received echo data and extracting zero Doppler data comprises: Transmitting waves based on antenna and waveform configuration; The hardware accelerator in the hardware platform performs 2D-FFT processing on the echo signal; After hardware acceleration processing, a signal matrix with a range dimension length of m and a Doppler dimension length of n is obtained; Acquire zero Doppler data.

4. The method for comprehensively distinguishing the type of living things in a cabin based on multiple features according to claim 3 is characterized in that: The signal matrix only retains signals where the Doppler value is zero.

5. The method for comprehensively distinguishing the type of living things in a cabin based on multiple features according to claim 4 is characterized in that: The DBF processing of the zero Doppler signals of a plurality of consecutive frames comprises: In the receiving channel, the zero Doppler signals of the first group of azimuth dimension channels and the first group of elevation dimension channels are selected for DBF processing respectively. The formula is as follows: in, is the DBF energy spectrum, is the steering vector, R is the input signal covariance, L is the number of channels, is the selected DBF angle.

6. The method for comprehensively distinguishing the type of living things in a cabin based on multiple features according to claim 5 is characterized in that: The judgment criteria include: A. When the in-cabin liveness detection radar is powered on, the measured results are obtained based on the above steps. , , and Features are input into the classifier for type discrimination, and the discrimination results and confidence are output, where the confidence is obtained by calculating the L2 norm of the measured features and fitting parameters; B. When the classification result of the classifier conflicts with the respiratory rate, the classification results obtained by each feature are sorted based on the weights and confidence levels of different classification features, and the classification result with the largest product of the two is output.

Images