A data-fingerprint-based millimeter wave radar data fast calculation method

Abstract

The application discloses a millimeter wave radar data rapid calculation method based on data fingerprints, relates to the technical field of millimeter wave radars, discloses steps of making a fingerprint library, fingerprint matching calculation, judgment result and the like, utilizes millimeter wave radar data fingerprints, performs fuzzy estimation on a calculation result, thereby screens data which is "effective" to the result, greatly reduces data entering a computing power module, and further improves the efficiency of the computing module.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of millimeter wave radar, more particularly, it relates to a millimeter wave radar data fast calculation method based on data fingerprint. BACKGROUND

[0002] With the localization of millimeter wave radar chips and the popularization of various scene radar chips, more and more edge computing gateway devices capable of supporting more computing power are needed, however, due to the influence of the cost of computing chips and chip control policies, it is difficult to bear the demand of all millimeter wave radar data processing computing power on domestic computing modules. Through analyzing the millimeter wave radar data calculation process and calculation results, it is found that most of the calculation results of the calculation module are non-positive results, that is, for the positive result response, these operations can undoubtedly be called "invalid" operations. Therefore, it is necessary to simplify or reduce these "invalid" operations to use limited operation resources for "valid" operations. SUMMARY

[0003] In view of the defects in the prior art, the purpose of the present application is to provide a millimeter wave radar data fast calculation method based on data fingerprint.

[0004] To achieve the above purpose, the present application provides the following technical scheme:

[0005] A millimeter wave radar data fast calculation method based on data fingerprint, comprising the following steps:

[0006] Step one: align the millimeter wave radar to the object to be tested, collect N frames of millimeter wave radar data, mark the time sequence X(t), mark the label corresponding to the time sequence as mark, wherein t≤N, X(t)={X(1), X(2), …, X(N)}, convert each column in the time sequence X(t) to use tensor representation, its shape is (Nchirp, Nfreq), wherein Nchirp represents that the millimeter wave radar data frame contains Nchirp millimeter wave radar transmission periods, Nfreq represents the sampling times in one millimeter wave radar transmission period, and X(t) can also be represented by tensor, its shape is (N, Nchirp, Nfreq), the fingerprint is calculated by using frequency energy spectrum, the calculated fingerprint is stored, and the construction of the fingerprint library is completed;

[0007] Step two: upload the data generated by the millimeter wave radar according to the frame, calculate the current data fingerprint by using the frequency energy spectrum, and calculate the similarity between the data fingerprint and the fingerprint in the fingerprint library;

[0008] Step three: use the Euclidean space distance method to calculate the spatial distance between the data fingerprint and the fingerprints in the fingerprint library, and select the multiple fingerprints with the shortest distance and the corresponding labels to determine whether the labels are valid;

[0009] Step four: if the label is valid, further operation is performed, and if the label is invalid, the processing is not continued.

[0010] Further, A, a sequence of data composed of a plurality of chirps (a data set composed of a millimeter wave radar from a transmitted signal to the completion of all signal reception) is obtained, and the frequency energy spectrum px of the sequence is obtained through a fast Fourier transform algorithm; B, all chirp data frames are summed and averaged to obtain a data average frame, and the data average frame is used as a template for a DWT (dynamic time warping) algorithm; C, the difference matrix of the chirp data frame and the template is obtained by using the DWT algorithm in turn, all the difference matrices are processed by summing and are marked as a, then w=a / the number of non-zero values in the difference matrix, and w is used as the weight of the frequency energy spectrum; D, the frequency energy spectrum and the energy spectrum weight w in the tensor are calculated in turn by using a sliding window algorithm; E, the frequency energy spectrum normalized by the weight is used as the fingerprint of the sequence, the calculated fingerprint is stored, and the construction of the fingerprint library is completed, the construction of the fingerprint library is an offline process, and a record in the fingerprint library includes the fingerprint dataSign of the data and the label mark corresponding to the fingerprint, steps B and C are used to obtain the frequency energy spectrum weight, and other ways can be used for fingerprint calculation, such as short-time power density spectrum, mel frequency cepstral coefficient, etc., which can be used alone or in combination, as long as the algorithm in step A is adjusted to the corresponding algorithm.

[0011] Compared with the prior art, the present application has the following beneficial effects:

[0012] The millimeter wave radar data fingerprint is used to perform fuzzy estimation on the calculation result, so as to screen the data effective for the result, greatly reduce the data entering the computing module, and further improve the efficiency of the computing module. BRIEF DESCRIPTION OF DRAWINGS

[0013] Figure 1 It is a flowchart of a millimeter wave radar data fast calculation method based on data fingerprint. DETAILED DESCRIPTION

[0014] REFERENCE Figure 1 A millimeter wave radar data fast calculation method based on data fingerprint, comprising the following steps:

[0015] Step one: align the millimeter wave radar to the object to be tested, collect N frames of millimeter wave radar data, mark the time sequence X(t), mark the label corresponding to the time sequence as mark, where t≤N, X(t)={X(1), X(2), …, X(N)}, convert each column in the time sequence X(t) to a tensor representation with shape (Nchirp, Nfreq), where Nchirp represents the number of millimeter wave radar transmission cycles contained in the millimeter wave radar data frame, Nfreq represents the number of samples in one millimeter wave radar transmission cycle, and X(t) can also be represented by a tensor with shape (N, Nchirp, Nfreq). Calculate the fingerprint using the frequency energy spectrum, store the calculated fingerprint, and complete the construction of the fingerprint library; after determining the general task, M will become a constant, and Flag is a label that can mark the attributes of the time sequence X(t), its value can be a string (used for classification tasks), a numerical value (used for physical index detection), the collected and labeled data can be directly stored in the database or file. Taking continuous millimeter wave radar personnel distance measurement as an example, a person stands at a distance of d meters from the millimeter wave radar, assuming N=8, M=1024, because the intermediate frequency data of the continuous wave radar can be represented by a complex number, so its corresponding X(t) contains 8*1024 complex numbers, and the corresponding time sequence is collected at a distance of 1.1m, 1.4m, 1.7m, and 2.0m from the millimeter wave radar, respectively, and the distance is taken as the label to form the time sequence data with label, step one: align the millimeter wave radar to the object to be tested, collect N frames of millimeter wave radar data, mark the time sequence X(t), mark the label corresponding to the time sequence as mark, where t≤N, X(t)={X(1), X(2), …, X(N)} converts each column in the time series X(t) into a tensor representation with shape (N chirp, N freq), where N chirp represents N chirp millimeter wave radar transmission cycles contained in a millimeter wave radar data frame, and N freq represents the number of samples in one millimeter wave radar transmission cycle. Similarly, X(t) can also be represented by a tensor with shape (N, N chirp, N freq). The fingerprint is calculated using the frequency energy spectrum, and the calculated fingerprint is stored to complete the construction of the fingerprint library. The fingerprint is calculated using the frequency energy spectrum, and the fingerprint is calculated using the frequency energy spectrum. The specific steps include the following: A, a sequence of continuous multiple chirps (a data set composed of millimeter wave radar from transmitting signals to all signal reception) is formed, and the frequency energy spectrum px of the sequence is obtained by fast Fourier transform algorithm; B, all chirp data frames are summed and averaged to obtain a data average frame, and the data average frame is used as a template for DWT (dynamic time warping) algorithm; C, the difference matrix of the chirp data frame and the template is calculated using the DWT algorithm in turn, and all the difference matrices are squared and summed and marked as a. Then w=a / number of non-zero values in the difference matrix, and w is used as the weight of the frequency energy spectrum; D, the frequency energy spectrum and the energy spectrum weight w in the tensor are calculated using the sliding window algorithm in turn; E, the frequency energy spectrum normalized by the weight is taken as the fingerprint of the sequence, and the calculated fingerprint is stored to complete the construction of the fingerprint library. The construction of the fingerprint library is an offline process, and a record in the fingerprint library contains the fingerprint dataSign of the data and the label mark corresponding to the fingerprint. Steps B and C are used to obtain the frequency energy spectrum weight, and other methods can also be used for fingerprint calculation, such as short-time power density spectrum, mel frequency cepstral coefficient, etc. They can be used alone or in combination, as long as the algorithm in step A is adjusted to the corresponding algorithm.

[0016] Step two: upload the data generated by the millimeter wave radar according to the frame, calculate the data fingerprint using the frequency energy spectrum, and calculate the similarity between the data fingerprint and the fingerprint in the fingerprint library;

[0017] Step three: use the Euclidean space distance method to calculate the spatial distance between the data fingerprint and the fingerprint in the fingerprint library, and select the shortest multiple fingerprints and the corresponding labels, and judge whether the labels are valid;

[0018] Step four: if the label is valid, further operation is performed, and if the label is invalid, the processing is not continued.

[0019] If it is a classification task, the nearest one is selected as the output, and if it is a physical index detection task, the nearest multiple labels can be averaged and output.

[0020] The above merely describes the preferred embodiments of the present application, and the protection scope of the present application is not limited to the above-mentioned embodiments. Any technical scheme falling within the concept of the present application shall fall within the protection scope of the present application. It should be noted that, for ordinary skilled persons in the art, some improvements and refinements without departing from the principles of the present application shall also be considered as falling within the protection scope of the present application.

[0021] The above describes one embodiment of the present application in detail, but the content described is merely the preferred embodiment of the present application, and cannot be considered as limiting the implementation scope of the present application. Any equivalent changes and improvements made within the scope of the present application shall still fall within the patent coverage scope of the present application.

Claims

1. A method for rapid calculation of millimeter-wave radar data based on data fingerprinting, characterized in that, Includes the following steps: Step 1: Point the millimeter-wave radar at the object to be tested and collect N frames of millimeter-wave radar data, denoted as time series X(t). Label the time series with the label "mark", where t≤N, X(t)={X(1),X(2),…,X(N)}. Convert each column of the time series X(t) into a tensor representation with shape (Nchirp,Nfreq), where Nchirp represents the number of millimeter-wave radar transceiver cycles in the millimeter-wave radar data frame, and Nfreq represents the number of samples within one millimeter-wave radar transceiver cycle. Similarly, X(t) can also be represented by a tensor with shape (N, Nchirp, Nfreq). Calculate the fingerprint using the frequency energy spectrum. The specific steps for calculating the fingerprint using the frequency energy spectrum are as follows: A. Assemble a sequence from the data of multiple consecutive chirps, and obtain the frequency energy spectrum of the sequence using the Fast Fourier Transform algorithm; B. Sum all chirp data frames and take the average value to obtain the average data frame. Use the average data frame as the template for the DWT algorithm. C. Use the DWT algorithm to calculate the difference matrix between the chirp data frame and the template. Sum the squares of all the difference matrices and label them as a. Then w = a / the number of non-zero values ​​in the difference matrix. w is used as the weight of the frequency energy spectrum. D. Calculate the frequency energy spectrum and energy spectrum weight w in the tensor using the sliding window algorithm in turn; E. Take the weighted normalized frequency energy spectrum as the fingerprint of this sequence segment; The calculated fingerprints are stored to complete the construction of the fingerprint database; Step 2: Assemble the data generated by the millimeter-wave radar into frames and upload them. Calculate the current data fingerprint using the frequency energy spectrum and perform similarity calculations between the data fingerprint and fingerprints in the fingerprint database. Step 3: Use the Euclidean spatial distance method to calculate the spatial distance between the data fingerprint and the fingerprints in the fingerprint database, select the fingerprints with the shortest distance and their corresponding labels, and determine whether the labels are valid; Step 4: If the label is valid, proceed with the calculation; if the label is invalid, stop processing.

Images