Machine learning two-wheeled vehicle seating detection method based on millimeter wave radar

Abstract

The invention discloses a machine learning two-wheeled vehicle seating detection method based on a millimeter-wave radar, and belongs to the technical field of vehicle seating detection, and the millimeter-wave radar can realize personnel target identification based on signal echo signal distribution and seating and leaving state maintenance of signal time sequence change at low cost. Through the above mode, the method provided by the invention makes full use of the characteristic that the millimeter-wave radar is sensitive to micro motion, solves the problem of a pain point of a pressure sensor that people do not distinguish, judges the existence of a target based on machine learning, introduces an occupied grid, and enables the seating and leaving detection to be more stable.

Description

Technical Field

[0001] The present invention relates to the technical field of vehicle seat occupancy detection, and particularly to a method for detecting the occupancy of a two-wheeled vehicle based on a millimeter-wave radar and machine learning. Background Art

[0002] In recent years, with the acceleration of the urbanization process and the increasingly serious problem of traffic congestion, two-wheeled electric vehicles have received extensive attention as a green and portable means of travel.

[0003] Existing two-wheeled electric vehicles often require users to manually control the power on and off using a key or a button. This method has problems such as inconvenient operation and power waste caused by forgetting to turn off the power. In recent years, a method for detecting seat occupancy and departure based on a pressure sensor to start or disconnect the power of an electric vehicle has also emerged.

[0004] However, this type of solution also has the following problems:

[0005] 1. High cost;

[0006] 2. False alarms caused by placing other sundries;

[0007] 3. Unstable measured pressure due to road bumps.

[0008] Based on this, the present invention designs a method for detecting the occupancy of a two-wheeled vehicle based on a millimeter-wave radar and machine learning to solve the above problems. Summary of the Invention

[0009] In view of the above-mentioned disadvantages of the prior art, the present invention provides a method for detecting the occupancy of a two-wheeled vehicle based on a millimeter-wave radar and machine learning.

[0010] To achieve the above objectives, the present invention is realized through the following technical solutions:

[0011] A method for detecting the occupancy of a two-wheeled vehicle based on a millimeter-wave radar and machine learning includes the following steps:

[0012] Step 1: Install a millimeter-wave radar inside the seat under the cushion. The millimeter-wave radar is used to collect target information in real time;

[0013] Step 2: While collecting the target information, the millimeter-wave radar demodulates the target information to obtain a spectrum in the distance dimension;

[0014] Step 3: The millimeter-wave radar extracts the phase of the target information according to the DACM algorithm, calculates the difference in phase between the front and rear frames, and outputs the phase time series difference value of the micro-motion detection frame;

[0015] Step 4: The millimeter-wave radar determines the presence of the target based on machine learning, inputs the 1D-FFT spectrum information into random tree training for prediction, uses the weights trained by the random tree to perform real-time presence inference, and outputs the inference judgment result;

[0016] Step 5: Determine whether the criteria for someone sitting on the vehicle are met according to the output result of the random tree model and the phase time series difference value of the micro-motion detection frame, and then perform power-on or power-off operations on the two-wheeler.

[0017] Furthermore, in the above Step 1, the target information is the personnel position information.

[0018] Furthermore, obtaining the spectrum in the range dimension includes the following steps:

[0019] A1. The millimeter-wave radar sends a frequency-modulated continuous pulse signal through a transmitting antenna, receives the reflected echo signal of the target through two receiving antennas, mixes the echo signal and then demodulates it to obtain two-channel ADC data;

[0020] A2. Perform 1D-FFT processing on the two-channel ADC data respectively to obtain the spectrum in the range dimension.

[0021] Furthermore, the transmitting antenna is a TX antenna.

[0022] Furthermore, the receiving antenna is an RX antenna.

[0023] Furthermore, a frequency-modulated continuous pulse signal contains N linearly frequency-modulated signals in one frame period.

[0024] Furthermore, the specific steps for implementing Step 3 are as follows:

[0025] B1. Extract the phase information corresponding to the target information on each frame of complex 1D-FFT spectrum to obtain the real part signal amplitude I(t) and the complex part signal amplitude Q(t);

[0026]

[0027] where A[t] is the amplitude, is the phase, S[t] is the 1D-FFT function, and j is the imaginary unit;

[0028] B2. Use the spectrum in the range dimension to first differentiate and then integrate the phase difference between the front and back frames to remove high-frequency noise while restoring the phase change information. The calculation formula is:

[0029]

[0030] where ω(t) is the discretized analog phase angular velocity, is the first derivative of I[t] with respect to t, is the first derivative of Q[t] with respect to t;

[0031] is obtained by restoring the integral of the phase angular velocity over N chirp signals within one frame of time, is the phase change information; t is the time series, Δt is the time series difference, ω is the phase angular velocity, and m is the upper limit of the time series;

[0032] B3. Use the difference method to calculate the difference in phase between the previous and current frames and output the phase time series difference value of the micro-motion detection frame. When this difference value is higher than the threshold, it indicates that there is micro-motion at the target distance.

[0033] Furthermore, the threshold depends on the seat environment, seat cushion thickness, and seat cushion material.

[0034] Furthermore, the specific steps implemented in step four are as follows:

[0035] C1. Sum up the 1D-FFT values of N chirp signals in one frame and take the average. After weighting this average, it becomes the amplitude, and take this amplitude as the 1D-FFT observed amplitude of this frame, and then remove the stationary clutter;

[0036] C2. Take the 1D-FFT mean and variance within the target distance range of the 1D-FFT observed amplitude as the feature input to the random tree model;

[0037] C3. The random tree model then outputs the inference and judgment result.

[0038] Furthermore, the method for judging whether someone is sitting on the two-wheeler in step five is as follows:

[0039] D1. The random tree model judges that there is a target;

[0040] D2. The phase time series difference value of the micro-motion detection frame is higher than the threshold;

[0041] When both conditions are met, it is judged that someone is sitting currently and the two-wheeler is powered on;

[0042] When any one of the two conditions is not met, it is judged that no one is sitting currently and the power supply of the two-wheeler is disconnected.

[0043] Compared with the prior art, the beneficial effects of the present invention are as follows: 1. The method provided by the present invention makes full use of the characteristics of the millimeter-wave radar being sensitive to micro-motion and solves the pain point problem of the pressure sensor being unable to distinguish between people and objects;

[0044] 2. The present invention uses a millimeter-wave radar with lower cost to detect the point cloud caused by the deformation of the seat due to sitting, and introduces an occupancy grid based on machine learning to judge the existence of the target, making the detection of sitting and leaving the seat more stable and avoiding the problem of unstable measured pressure caused by road bumps. BRIEF DESCRIPTION OF THE DRAWINGS

[0045] In order to more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the following will briefly introduce the drawings required for use in the description of the embodiments or the prior art. Obviously, the drawings in the following description are only some embodiments of the present invention, and those of ordinary skill in the art can obtain other drawings based on these drawings without creative efforts.

[0046] Figure 1 It is a flowchart of a method for detecting the sitting of a two-wheeled vehicle based on machine learning using a millimeter-wave radar according to the present invention. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0047] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the following will clearly and completely describe the technical solutions in the embodiments of the present invention with reference to the drawings in the embodiments of the present invention. Obviously, the described embodiments are some, but not all, of the embodiments of the present invention. All other embodiments obtained by those of ordinary skill in the art based on the embodiments of the present invention without creative efforts fall within the scope of protection of the present invention.

[0048] Embodiment 1: In some embodiments, please refer to Figure 1 in the accompanying drawings of the specification, a method for detecting the sitting of a two-wheeled vehicle based on machine learning using a millimeter-wave radar includes the following steps:

[0049] Step 1: Install the millimeter-wave radar inside the seat under the cushion. The millimeter-wave radar is used to collect target information in real time, and the target information is the personnel position information.

[0050] Step 2: While collecting the target information, the millimeter-wave radar demodulates the target information to obtain the spectrum in the range dimension.

[0051] Obtaining the spectrum in the range dimension includes the following steps:

[0052] A1. The millimeter-wave radar sends a frequency-modulated continuous pulse signal through a TX (Transmitter) antenna, receives the reflected echo signal of the target through two RX (Receiver) antennas, mixes the echo signal and then demodulates it to obtain two-channel ADC (Analog to Digital Converter) data.

[0053] One frame period of the frequency-modulated continuous pulse signal contains N chirp signals; N > 0;

[0054] A2. Perform 1D-FFT (One Dimensional Fast Fourier Transformation) processing on the two-channel ADC data respectively to obtain the spectrum in the range dimension.

[0055] Step 3: The millimeter-wave radar extracts the phase of the target information according to the DACM algorithm (Digital Algorithm for Carrier Modulation), calculates the difference in phase between the previous and current frames, and outputs the difference value of the micro-motion detection frame phase time series;

[0056] The specific implementation steps are as follows:

[0057] B1. Extract the phase information corresponding to the target information on the 1D-FFT spectrum of each frame to obtain the real part signal amplitude I(t) and the complex part signal amplitude Q(t);

[0058]

[0059] where A[t] is the amplitude, is the phase, S[t] is the 1D-FFT function, and j is the imaginary unit;

[0060] B2. Use the spectrum in the range dimension to first differentiate and then integrate the phase difference between the previous and current frames to remove high-frequency noise and restore the phase change information. The calculation formula is:

[0061]

[0062] ω(t) is the discretized analog phase angular velocity; is the first derivative of I[t] with respect to t, is the first derivative of Q[t] with respect to t;

[0063]

[0064] where, is obtained by integrating the phase angular velocity on N chirp signals within one frame time, is the phase change information, t is the time series, Δt is the time series difference, ω is the phase angular velocity, and m is the upper limit of the time series.

[0065] B3. Use the difference method to calculate the difference in phase between the previous and current frames and output the difference value of the micro-motion detection frame phase time series. Here, when this difference value is higher than a certain threshold, it indicates that there is micro-motion in the target distance.

[0066] The target range (Range Bin) is an important concept in radar signal processing. It refers to the discretized range cells in the radar detection space. In a radar system, the distance to a target is calculated by measuring the time it takes for an electromagnetic wave to travel from the radar to the target and back. Since the propagation speed of electromagnetic waves in space is constant, the distance to the target can be determined through time-distance conversion.

[0067] The threshold needs to be adjusted according to different seat environments, which is related to the seat cushion thickness, seat cushion material, etc., and requires some fine-tuning.

[0068] Step 4: The millimeter-wave radar determines the presence of a target based on machine learning. It inputs the 1D-FFT spectrum information into random tree training for prediction, uses the weights trained by the random tree to perform real-time presence inference, and outputs the inference judgment result.

[0069] The specific implementation steps are as follows:

[0070] C1. Sum the 1D-FFT values of N chirp signals in a frame and take the average. Weight this average to obtain the amplitude, and take this amplitude as the 1D-FFT observation amplitude of this frame. Then remove the stationary clutter.

[0071] C2. Take the 1D-FFT mean and variance within the target range of the 1D-FFT observation amplitude as features and input them into the random tree model.

[0072] C3. The random tree model then outputs the inference judgment result.

[0073] Step 5: The millimeter-wave radar determines whether it meets the standard of someone sitting based on the output result of the random tree model and the phase time series difference value of the micro-motion detection frame.

[0074] The method for judging whether someone is sitting on a two-wheeler is as follows:

[0075] D1. The random tree model determines that there is a target.

[0076] D2. The phase time series difference value of the micro-motion detection frame is higher than a certain threshold.

[0077] When both conditions are met, it is determined that someone is sitting currently, and the two-wheeler is powered on.

[0078] When any one of the two conditions is not met, it is determined that no one is sitting currently, and the power supply of the two-wheeler is disconnected.

[0079] The method provided by the present invention makes full use of the characteristic that the millimeter-wave radar is sensitive to micro-motions, can sensitively identify people or objects, and solves the problem that the pressure sensor cannot distinguish between people and objects.

[0080] The present invention uses a millimeter-wave radar with lower cost to detect the point cloud caused by the deformation of the seat during seating, and introduces an occupancy grid based on machine learning to judge the presence of the target, making the detection of seating and leaving the seat more stable and avoiding the problem of unstable measured pressure caused by road bumps.

[0081] The above embodiments are only used to illustrate the technical solutions of the present invention, rather than to limit it; although the present invention has been described in detail with reference to the foregoing embodiments, those of ordinary skill in the art should understand that they can still modify the technical solutions described in the foregoing embodiments, or perform equivalent replacements on some of the technical features; and these modifications or replacements will not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of the present invention.

Claims

1. A machine learning two-wheeled vehicle seat detection method based on millimeter wave radar, characterized in that: The following steps are involved: Step 1: Install the millimeter wave radar inside the seat under the cushion. The millimeter wave radar is used to collect target information in real time; Step 2: The millimeter wave radar demodulates the target information while collecting it to obtain the frequency spectrum of the distance dimension; Step 3: The millimeter-wave radar extracts the phase of the target information according to the DACM algorithm, calculates the difference between the phases of the previous and next frames, and outputs the phase timing difference value of the micro-motion detection frame; Step 4: The millimeter-wave radar determines the existence of the target based on machine learning, inputs the 1D-FFT spectrum information into the random tree training for prediction, uses the weights trained by the random tree to perform real-time existence reasoning, and outputs the reasoning judgment results; Step 5: Determine whether the occupancy standard is met based on the output results of the random tree model and the phase timing difference value of the micro-motion detection frame, and then power on or off the two-wheeled vehicle.

2. The machine learning two-wheeled vehicle seat detection method based on millimeter wave radar according to claim 1 is characterized in that: In the step 1, the target information is personnel location information.

3. The machine learning two-wheeled vehicle seat detection method based on millimeter wave radar according to claim 1 is characterized in that: Obtaining the spectrum of the distance dimension includes the following steps: A1. The millimeter-wave radar sends a frequency-modulated continuous pulse signal through a transmitting antenna, and receives the reflected echo signal of the target through two receiving antennas. The echo signal is mixed and demodulated to obtain two ADC data. A2. Perform 1D-FFT processing on the two channels of ADC data respectively to obtain a frequency spectrum of the distance dimension.

4. The machine learning two-wheeled vehicle seat detection method based on millimeter wave radar according to claim 3 is characterized in that: The transmitting antenna is a TX antenna.

5. The machine learning two-wheeled vehicle seat detection method based on millimeter wave radar according to claim 3 is characterized in that: The receiving antenna is a RX antenna.

6. The machine learning two-wheeled vehicle seat detection method based on millimeter wave radar according to claim 3 is characterized in that: The frequency modulated continuous pulse signal contains N linear frequency modulated signals in one frame period.

7. The machine learning two-wheeled vehicle seat detection method based on millimeter wave radar according to claim 1 is characterized in that: The specific steps to implement step three are: B1. Extract the phase information corresponding to the target information on each frame of the complex 1D-FFT spectrum to obtain the real part signal amplitude I(t) and the complex part signal amplitude Q(t); Where A[t] is the amplitude, is the phase, S[t] is the 1D-FFT function, and j is the imaginary unit; B2. Use the spectrum of the distance dimension to differentiate and then integrate the phase difference between the previous and next frames, remove high-frequency noise and restore the phase change information. The calculation formula is: Where, ω(t) is the phase angular velocity of the discretized simulation; is the first-order derivative of I[t] with respect to t, is the first-order derivative of Q[t] with respect to t; is the phase change information, t is the time series, Δt is the time series difference, ω is the phase angular velocity, and m is the upper limit of the time series; B3. Use the differential method to calculate the difference between the previous and next frame phases and output the micro-motion detection frame phase timing difference value. When the difference value is higher than the threshold value, it indicates that there is micro-motion at the target distance.

8. The machine learning two-wheeled vehicle seat detection method based on millimeter wave radar according to claim 7 is characterized in that: The threshold value depends on the vehicle seat environment, seat cushion thickness and seat cushion material.

9. The machine learning two-wheeled vehicle seat detection method based on millimeter wave radar according to claim 1 is characterized in that: The specific steps to implement step 4 are: C1. The 1D-FFT values ​​of N linear frequency modulation signals in a frame are summed up and averaged, and the average value is weighted to become an amplitude, and the amplitude is taken as the 1D-FFT observation amplitude of the frame, and then the stationary clutter is removed; C2, take the 1D-FFT mean and variance within the target distance range of the 1D-FFT observation amplitude as features to input into the random tree model; C3. The random tree model then outputs the inference judgment result.

10. The machine learning two-wheeled vehicle seat detection method based on millimeter wave radar according to claim 9 is characterized in that: The method for judging whether a two-wheeled vehicle has a seat in step 5 is as follows: D1, random tree model judges that there is a target; D2, the phase timing difference value of the micro-motion detection frame is higher than the threshold; When both conditions are met, it is determined that someone is currently sitting in the car, and the two-wheeled vehicle is powered on; When any of the two conditions is not satisfied, it is determined that no one is currently seated, and the power supply of the two-wheeled vehicle is disconnected.

Images