Target detection method and system and storage medium

Abstract

The invention provides a target detection method and system and a storage medium, and the method comprises the steps: collecting original data in real time in a preset measurement range of a vehicle, and carrying out the primary processing of the original data, and obtaining scene data; performing target occurrence probability prediction on the scene data based on a target detection model to obtain target prediction data; performing prediction result judgment on the target prediction data based on a fixed threshold value, so as to obtain target point detection data according to a judgment result; and performing condensation processing on all the target point detection data to obtain final target detection data. According to the method provided by the invention, hardware resources are effectively reduced, high-performance real-time target monitoring is realized, and target detection can be well carried out in a low signal-to-noise ratio and strong clutter environment.

Description

Technical Field

[0001] This application belongs to the technical field of radar target detection, and specifically relates to a target detection method, system and storage medium. Background Art

[0002] As an advanced wireless detection technology, millimeter-wave radar technology has been widely applied and developed in both military and civilian fields since the mid-20th century. With the continuous progress of radar technology and the reduction of costs, millimeter-wave radar has gradually entered the civilian market, especially in the field of automotive safety, where significant achievements have been made.

[0003] Due to various factors such as significant multipath effects, strong environmental clutter, strong electromagnetic interference, poor electromagnetic compatibility between devices, and strong noise when applying actual detection algorithms, existing methods cannot obtain good detection performance in an environment with low signal-to-noise ratio and strong clutter. For example, under a constant false alarm rate, the false alarm prediction is high and the detection prediction is low, resulting in the inability to directly and well apply the detection algorithm to the actual situation. Due to limited computing power, more complex algorithms cannot meet the requirements of real-time monitoring. DDMA (Doppler Division Multiple Access) radar sacrifices velocity ambiguity, which affects the detection effect. Currently, existing coherent and non-coherent detection algorithms cannot well accumulate the energy of DDMA signals, and the detection effect is not good.

[0004] There will be a problem of target aliasing in DDMA radar. Currently, existing coherent and non-coherent detection algorithms cannot well distinguish multiple virtual images generated by different targets on the range-Doppler map (RDM). Different targets will be aliased, resulting in a decline in detection performance. Summary of the Invention

[0005] To solve the above technical problems, this application proposes a target detection method, system and storage medium, aiming to solve the technical problem of low target detection performance in an environment with low signal-to-noise ratio and strong clutter in the prior art. It effectively solves the problems of low detection probability of weak targets and high false alarm probability in a strong clutter environment in the prior art.

[0006] Specifically, this application proposes a target detection method, including:

[0007] Real-time collect raw data within a preset measurement range of the vehicle, and perform preliminary processing on the raw data to obtain scene data.

[0008] Based on a target detection model, predict the probability of target appearance for the scene data to obtain target prediction data.

[0009] Based on a fixed threshold, make a decision on the prediction result of the target prediction data to obtain target point detection data according to the decision result.

[0010] Further, all the target point detection data are subjected to condensation processing to obtain the final target detection data.

[0011] In the above technical solution, by initially processing the raw data collected in real time, the timeliness of the target detection process is ensured, and the efficiency of the target detection process is improved through the initial processing of the raw data. The probability of the target appearance is predicted through the target detection model, thereby improving the accuracy of the target detection, reducing the phenomena of false detection and missed detection, and ensuring the reliability and accuracy of the target detection. The prediction result of the target prediction data is judged by a fixed threshold value, reducing the false alarm situation. By subjecting the target point detection data to condensation processing, the error or noise interference in the target detection process is reduced, thereby improving the reliability and stability of the target detection result.

[0012] As an implementation manner, the initial processing at least includes energy accumulation processing and pulse compression processing; wherein, the energy accumulation processing includes:

[0013] The raw data are formed into multi-dimensional matrix data; wherein, the multi-dimensional matrix data at least include a radar antenna channel dimension, a time dimension, and a pulse dimension; the multi-dimensional matrix data are sliced based on the radar antenna channel dimension to obtain initial sliced matrix data; a first preset number of the initial sliced matrix data are obtained to obtain target sliced matrix data based on the first preset number of the initial sliced matrix data.

[0014] Forming the raw data into multi-dimensional matrix data provides a clear data structure for initially processing the raw data, enabling flexible initial processing of the raw data through the radar antenna channel dimension, time dimension, and pulse dimension of the multi-dimensional matrix data, and improving the efficiency and flexibility of data processing. By slicing the multi-dimensional matrix data, it is possible to precisely analyze the data of different times, pulses, and radar antenna channels, effectively isolate the raw data of different dimensions, reduce noise interference, and improve the accuracy of target detection. The signal-to-noise ratio of the target detection is improved through the energy accumulation processing, enhancing the target detection performance.

[0015] Further, the pulse compression processing includes:

[0016] The target sliced matrix data are compressed based on the pulse dimension; a Fourier transform is performed on the compressed target sliced matrix data based on the time dimension to obtain the scene data.

[0017] The range resolution of the target detection process is improved through the pulse compression processing. By performing Fourier transform on the target slice matrix data after the compression processing, it is possible to accurately extract and analyze data in different frequency ranges, thereby improving the time resolution of the target. Through the pulse compression processing, it is possible to more accurately locate the range and speed of the target, ensuring the tracking accuracy of the target detection. The pulse compression processing can effectively improve the detection signal of the target, reduce the influence of stray noise and interference signals, improve the detection ability of weak echo signals, and through Fourier transform, it is possible to distinguish effective signals from interference signals, further improving the anti-interference ability and enhancing the anti-interference ability of the target detection process in complex environments.

[0018] Further, the target detection model at least includes a feature extraction model, a target prediction model, and a complex convolution model; the detection of the scene data based on the target detection model includes:

[0019] Extracting target semantic features from the scene data through the feature extraction model; predicting the probability of target appearance for the target semantic features through the target prediction model to obtain a target prediction matrix; obtaining the target prediction matrix of a second preset value through the complex convolution model to obtain the target prediction data based on the target prediction matrix of the second preset value.

[0020] Extracting target semantic features from the scene data through the feature extraction model enables the target detection process to better adapt to complex and variable scene data, enhancing the robustness of the target detection. Predicting the probability of target appearance for the semantic features through the target prediction model enables effective evaluation of the existence probability of the target, thereby obtaining a target prediction matrix, effectively improving the accuracy of target detection and reducing the occurrence of false detections and missed detections. Obtaining target prediction data through the complex convolution model enables processing of complex-valued data, enhancing the processing ability of complex data, and obtaining target prediction data through the target prediction matrix of the second preset value improves the accuracy of target prediction.

[0021] Further, the pre-training of the target detection model includes:

[0022] Detecting historical scene data through a pre-established target detection model to obtain historical target prediction data.

[0023] Obtaining historical target ground truth data to calculate the weight coefficient of the target detection model based on the historical target ground truth data and the historical target prediction data.

[0024] Optimizing the target detection model based on the weight coefficient to obtain the final target detection model.

[0025] Calculate the weight coefficient of the object detection model based on the historical target true value data and the historical target prediction data, so as to effectively optimize the object detection model, improve the accuracy of the object detection model, and improve the reliability and stability of the object detection.

[0026] Further, the decision-making on the object prediction data by a fixed threshold includes:

[0027] Obtain the predicted value of each point in the object prediction data by using a preset processing algorithm based on the object prediction data.

[0028] Sort the predicted values, and obtain the fixed threshold from the sorted predicted values based on a preset false alarm rate.

[0029] Judge the predicted values based on the fixed threshold by using a preset decision rule to obtain the object point detection data according to the decision result.

[0030] By setting a fixed threshold to screen the predicted values, false detections can be effectively reduced. By sorting the predicted values and selecting the fixed threshold, the false alarm rate can be effectively controlled, thereby improving the accuracy of object detection. By sorting and using a preset rule to judge the predicted values, the calculation of data can be effectively reduced, thereby improving the efficiency of obtaining object point detection data. Judging the predicted values based on the fixed threshold ensures the consistency and reliability of the decision result.

[0031] Further, the agglomeration processing of all the object point detection data includes:

[0032] Obtain the object point detection data with the largest amplitude based on all the object point detection data; agglomerate all the object point detection data to the position of the object point detection data with the largest amplitude.

[0033] By agglomerating the object point detection data with the object point detection data with the largest amplitude, all the object point detection data can be fused, the accuracy of object position detection can be improved, the error caused by noise or other factors can be reduced, and the reliability of the detection result can be ensured. The consistency and stability of object detection are enhanced.

[0034] Further, after obtaining the final object detection data, it further includes:

[0035] Obtain the object detection data of the third preset value.

[0036] Calculate and obtain the number of repetitions of the object detection data at the same coordinate based on the object detection data.

[0037] Determine whether the number of repetitions is greater than a preset number threshold. If so, determine that the target detection data exists; otherwise, determine that the target detection data is abnormal and discard the abnormal target detection data.

[0038] By calculating the number of repetitions of the target detection data, it is possible to effectively identify misdetected and invalid detection data. When the number of repetitions is less than the preset number threshold, it is determined that the target detection data is abnormal, which may be due to errors, noise, or data deviation. By discarding the abnormal target detection data, abnormal data can be effectively filtered out, ensuring the accuracy and reliability of the final target detection data. This improves the robustness and consistency of target detection.

[0039] Based on the same inventive concept, the present application also proposes a system for a target detection method, which includes:

[0040] A data acquisition module for real-time collecting raw data within a preset measurement range of the vehicle.

[0041] A processing module for preliminarily processing the raw data to obtain scene data.

[0042] A prediction module for predicting the probability of the appearance of a target based on a target detection model for the scene data to obtain target prediction data.

[0043] A decision module for making a decision on the target prediction data through a fixed threshold to obtain target point detection data according to the decision result.

[0044] And a condensation module for condensing all the target point detection data to obtain the final target detection data.

[0045] Based on the same inventive concept, the present application also proposes a computer-readable storage medium storing computer-executable instructions that can be read and executed by a domain controller to perform the target detection method.

[0046] Compared with the prior art, the present application has at least the following beneficial effects:

[0047] The object detection method proposed in this application effectively solves the technical problem of low object detection performance in the prior art under the environment of low signal-to-noise ratio and strong clutter. It effectively solves the problems of detecting small and weak objects and high probability of missing alarms in a strong clutter environment. By preliminarily processing the raw data collected in real time, the object detection process has timeliness, and the efficiency of the object detection process is improved through the preliminary processing of the raw data. By predicting the probability of the appearance of an object through an object detection model, the accuracy of object detection is improved, the phenomena of false detection and missed detection are reduced, and the reliability and accuracy of object detection are ensured. By using a fixed threshold to judge the prediction result of the object prediction data, the situation of false alarms is reduced. By aggregating the object point detection data, the error or noise interference in the object detection process is reduced, thereby improving the reliability and stability of the object detection result. Description of the Drawings

[0048] Figure 1 is a flowchart of the object detection method shown in an embodiment of the present application.

[0049] Figure 2 is a schematic diagram of the object detection system shown in an embodiment of the present application. Detailed Embodiments

[0050] Next, the technical solutions in the embodiments of the present application will be clearly and completely described in conjunction with the drawings in the embodiments of the present application. Obviously, the described embodiments are only a part of the embodiments of the present application, rather than all the embodiments. Based on the embodiments in the present application, all other embodiments obtained by those of ordinary skill in the art without creative efforts shall fall within the protection scope of the present application.

[0051] It should be noted that the terms "first", "second", etc. in the specification and claims of the present application and the above drawings are used to distinguish similar objects, and do not necessarily need to describe a specific order or sequence. It should be understood that such data can be interchanged under appropriate circumstances so that the embodiments of the present application described herein can be implemented in an order other than those illustrated or described herein. In addition, the terms "comprising" and "having" and any variations thereof are intended to cover non-exclusive inclusion. For example, a process, method, system, product, or server that includes a series of steps or units does not necessarily need to be limited to those steps or units clearly listed, but may include other steps or units that are not clearly listed or are inherent to these processes, methods, products, or devices.

[0052] Embodiment 1:

[0053] Please refer to Figure 1 , and this object detection method mainly includes steps S1 to S4.

[0054] Among them, step S1 includes: collecting raw data in real time within a preset measurement range of the vehicle, and performing preliminary processing on the raw data to obtain scene data. The raw data can be collected in real time within the preset measurement range of the vehicle by using a MIMO (Multiple Input Multiple Output) radar. The MIMO radar is generally a vehicle-mounted millimeter-wave radar, and the radar frequency is between 20 and 100 GHz. The MIMO radar means that the radar has multiple transmit and receive antennas, that is, multiple transmit antennas and multiple receive antennas work simultaneously. The two types of antennas are separated, and the signals are made as orthogonal as possible through a modulation method between the antennas, thereby improving the signal-to-noise ratio. Those skilled in the art can set the preset measurement range of the vehicle according to the actual situation. For example, the preset measurement range of the vehicle can be set to 50 meters. The preliminary processing at least includes pulse compression processing and energy accumulation processing. The energy accumulation processing mainly includes accumulating the energy of the target reflection signal in the raw data together. The pulse compression processing mainly includes compressing and Fourier-transforming the raw data based on the dimensions in the raw data. The scene data is mainly RDM (Range Doppler Map).

[0055] Step S2 includes: predicting the probability of target appearance for the scene data based on a target detection model to obtain target prediction data. The target detection model is mainly a data-driven coherent target detection model based on Doppler multiple access radar technology. The Doppler multiple access radar technology is a new type of radar technology that combines multiple access technology and the Doppler effect, aiming to improve the radar system's ability to simultaneously detect and separate multiple targets. The data-driven coherent target detection algorithm realizes the efficient and accurate detection of targets through the analysis of radar echo data, especially the target recognition in a multi-target and complex environment. Among them, in the data-driven coherent target detection model, the scene data is mainly analyzed through deep learning algorithms and machine learning algorithms.

[0056] Step S3 includes: making a prediction result decision on the target prediction data based on a fixed threshold, so as to obtain target point detection data according to the decision result. The fixed threshold is mainly a prediction threshold value set based on a constant false alarm rate, and the prediction result decision is made on the target prediction data based on the prediction threshold value. For example, the false alarm rate can be set to 10 -6 、10 -4 、10 -3 . Those skilled in the art can set the prediction threshold value with different constant false alarm rates according to the actual situation, and it is not limited to this.

[0057] Moreover, step S4 includes: performing condensation processing on all the target point detection data to obtain the final target detection data. The condensation processing mainly takes the position of the target detection point with the largest amplitude in the target point detection data as the condensation point position, condenses all the target point detection data to this condensation point position, and fuses to obtain the final target detection data.

[0058] In some embodiments, the preliminary processing at least includes energy accumulation processing and pulse compression processing; wherein, the energy accumulation processing includes:

[0059] Constructing the original data into multi-dimensional matrix data; wherein, the multi-dimensional matrix data at least includes a radar antenna channel dimension, a time dimension, and a pulse dimension.

[0060] Performing slicing processing on the multi-dimensional matrix data based on the radar antenna channel dimension to obtain initial sliced matrix data.

[0061] Obtaining the initial sliced matrix data of a first preset value, so as to obtain target sliced matrix data based on the initial sliced matrix data of the first preset value.

[0062] For example, the echo on the k-th receiving antenna after the echo transmitted by the j-th transmitting antenna of the MIMO radar passes through target and scene reflection and the influence of noise can be expressed as:

[0063]

[0064] Wherein, is the target reflected echo, n(t) is clutter, noise, and the sum of signals of other target reflected echoes, which is uniformly regarded as noise. For this MIMO radar, M pulses will be repeatedly transmitted within one frame, J transmitting antennas will simultaneously send JM pulses, and after passing through environment and target reflection, MK echoes will be obtained by K receiving antennas. The echoes within one pulse form an echo vector, and the target sliced matrix data obtained from M echo vectors is:

[0065]

[0066] Optionally, the pulse compression processing includes:

[0067] Performing compression processing on the target sliced matrix data based on the pulse dimension; performing Fourier transform on the target sliced matrix data after compression processing based on the time dimension to obtain the scene data.

[0068] Among them, the formula that can be adopted is:

[0069] X RDM =DFT D＝0 (Pc(Xori )) ∈ C M×N ;

[0070] Among them, DFT D＝0 represents performing a Fourier transform on each column of data in the matrix, and Pc represents performing compression processing on each row of data in the matrix data.

[0071] The obtained scene data can be:

[0072]

[0073] Optionally, the target detection model at least includes a feature extraction model, a target prediction model, and a complex convolution model; detecting the scene data based on the target detection model includes:

[0074] Extracting target semantic features from the scene data through the feature extraction model.

[0075] Predicting the probability of target appearance for the target semantic features through the target prediction model to obtain a target prediction matrix.

[0076] Obtaining the target prediction matrix of a second preset value through the complex convolution model to obtain the target prediction data based on the target prediction matrix of the second preset value.

[0077] Among them, the feature extraction model can mainly be a 1D-CCASPP (One-Dimensional Complex Circular Atrous Convolutional Spatial Pyramid Pooling) model. The 1D-CCASPP model combines complex convolution models of complex convolution, atrous convolution, and convolutional pyramid pooling. It mainly consists of a complex circular atrous convolution layer, a regularization layer, an activation layer, and a 1×1 convolution layer. Among them, the scene data can mainly pass through a complex circular atrous convolution layer, and the convolutional kernel has a configurable dilation rate, enabling the receptive field to be enlarged through atrous operations while maintaining a low computational cost. The features are standardized through the regularization layer to reduce instability during training. The regularized feature map will pass through an activation function to increase the non-linear expression ability of the network. In the convolutional pyramid pooling layer, multi-scale pooling is used to further extract features of different scales to obtain a richer feature representation. Finally, 1×1 convolution is used for feature fusion or channel adjustment to output the final target semantic features.

[0078] Optionally, the pre-trained target detection model includes:

[0079] Detect the historical scenario data through a pre-established target detection model to obtain historical target prediction data.

[0080] Obtain historical target ground truth data to calculate the weight coefficient of the target detection model based on the historical target ground truth data and the historical target prediction data. Optimize the target detection model based on the weight coefficient to obtain the final target detection model.

[0081] Among them, the weight coefficient can be defined as θ i , and the historical target prediction data output according to the target detection model is The obtained historical target ground truth data is Y. The distance difference between the historical target prediction data and the historical target ground truth data can be calculated according to the BCE (Binary Cross-Entropy Loss, binary cross-entropy) loss function. The calculation formula is:

[0082] L(Y * ,Y) = BCE(Y * ,Y)

[0083]

[0084] Calculate the gradient of the loss function with respect to the weight coefficient based on the distance difference Update the weight coefficient of the target detection model based on the gradient of the weight coefficient. The updated weight coefficient is where η is the set optimization rate. By performing multiple iterations to make the weight coefficient converge, the data-driven coherent target detection network based on Doppler frequency division multiple access radar technology can perform coherent detection on the targets in the scenario data.

[0085] Optionally, the decision on the target prediction data by a fixed threshold includes:

[0086] Based on the target prediction data, use a preset processing algorithm to obtain the prediction value of each point in the target prediction data.

[0087] Sort the prediction values, and obtain a fixed threshold from the sorted prediction values based on a preset false alarm rate.

[0088] Based on the fixed threshold, use a preset decision rule to determine the prediction values, so as to obtain target point detection data according to the decision result.

[0089] Among them, the preset processing algorithm can mainly be:

[0090]

[0091] The preset false alarm rate can mainly be the false alarm rate set when there is no target point prediction data within the prediction range. For example, based on the preset false alarm rate P fa , the predicted value of each point in the target prediction data is y (m,n) , sort the M*N predicted values in descending order, and then use the predicted values of int(MNP fa ) points as the fixed threshold τ B , where int() is rounding down.

[0092] Optionally, the agglomeration processing of all the target point detection data includes:

[0093] Obtain the target point detection data with the largest amplitude based on all the target point detection data; agglomerate all the target point detection data to the position of the target point detection data with the largest amplitude.

[0094] Among them, mainly by setting the set of target points before agglomeration as is Among them, each target point before agglomeration has a corresponding amplitude The position of each target point detection data with the largest amplitude uniquely corresponds to a connected region, that is

[0095] Optionally, after obtaining the final target detection data, it further includes:

[0096] Obtain the target detection data of the third preset value; calculate the number of repetitions of the target detection data appearing at the same coordinates based on the target detection data; determine whether the number of repetitions is greater than the preset number threshold, if so, determine that the target detection data exists, otherwise, determine that the target detection data is abnormal, and discard the abnormal target detection data.

[0097] For example, assuming that the third preset value is K, the number of times Q that the target appears at the same coordinates (r, d) through the K target detection data, and set the detection repetition number threshold to Q x , when Q≥Q x , it is determined that the target detection data exists, otherwise, it is determined that the target detection data is a false alarm, and the target detection data is discarded.

[0098] Embodiment 2:

[0099] Please refer to Figure 2 , this application also proposes a system using the target detection method described in Embodiment 1, which mainly includes: a data acquisition module, a processing module, a prediction module, a decision module, and an agglomeration module.

[0100] Among them, the data acquisition module is used to collect raw data in real time within the preset measurement range of the vehicle. A MIMO radar can be used to collect raw data in real time within the preset measurement range of the vehicle. The MIMO radar refers to a radar with multiple transmit and receive antennas. Those skilled in the art can set the preset measurement range of the vehicle according to the actual situation. For example, the preset measurement range of the vehicle can be set to 50 meters.

[0101] The processing module is used to perform preliminary processing on the raw data to obtain scene data. The preliminary processing can at least include energy accumulation processing and pulse compression processing. The energy accumulation processing mainly constitutes the raw data into multi-dimensional matrix data. Among them, the multi-dimensional matrix data at least includes the radar antenna channel dimension, the time dimension, and the pulse dimension. Based on the radar antenna channel dimension, slice processing is performed on the multi-dimensional matrix data to obtain initial slice matrix data. Obtain the initial slice matrix data of the first preset value, so as to obtain target slice matrix data based on the initial slice matrix data of the first preset value. The pulse compression processing mainly includes performing compression processing on the target slice matrix data based on the pulse dimension; performing Fourier transform on the compressed target slice matrix data based on the time dimension to obtain the scene data

[0102] The prediction module is used to predict the probability of the appearance of a target based on the target detection model for the scene data to obtain target prediction data. Among them, the target detection model can include a feature extraction model. The feature extraction module can be a one-dimensional complex circular dilated convolutional pyramid model. Those skilled in the art can select other feature extraction models according to the actual situation. For example, a circular dilated convolutional model can be used to replace the one-dimensional complex circular dilated convolutional pyramid model.

[0103] The decision module is used to make a decision on the target prediction data through a fixed threshold to obtain target point detection data according to the decision result. Among them, the fixed threshold can be mainly selected from the target prediction data through a preset false alarm rate, and a decision is made on the target prediction data based on the fixed threshold to obtain the target point detection data.

[0104] In addition, the aggregation module is used to perform aggregation processing on all the target point detection data to obtain the final target detection data. The aggregation processing can mainly be to aggregate all the target point prediction data in the target point detection data to the position of the target point with the largest amplitude to obtain the target detection data.

[0105] Embodiment 3:

[0106] The present application also proposes a domain controller, including a computer-readable storage medium, and the computer-readable storage medium includes:

[0107] The computer-readable storage medium stores computer-executable instructions.

[0108] When the computer-executable instructions are executed by a control processor, the target detection method described in the first embodiment is implemented.

[0109] In the computer-readable storage medium, it can be implemented in whole or in part by software, hardware, firmware, or any combination thereof. When implemented using software, it can be implemented in whole or in part in the form of a computer program product. The computer program product includes one or more computer instructions. When the computer program instructions are loaded and executed on a computer, the processes or functions described in the embodiments of the present application are generated in whole or in part. The computer can be a general-purpose computer, a special-purpose computer, a computer network, or other programmable devices. The computer instructions can be stored in a computer-readable storage medium, or transmitted from one computer-readable storage medium to another computer-readable storage medium. For example, the computer instructions can be transmitted from one website, computer, server, or data center to another website, computer, server, or data center by wire (such as coaxial cable, optical fiber, digital subscriber line) or wirelessly (such as infrared, wireless, microwave, etc.). The computer-readable storage medium can be any available medium that a computer can access or a data storage device such as a server or data center that includes one or more integrated available media. The available medium can be a magnetic medium (such as a floppy disk, hard disk, magnetic tape), an optical medium (such as a DVD), or a semiconductor medium (such as a solid-state disk (SSD)).

[0110] In summary, the target detection method proposed in this application effectively solves the technical problem of low target detectability in the prior art under the environment of low signal-to-noise ratio and strong clutter. It effectively solves the problem of high probability of missed alarms when detecting small and weak targets and in a strong clutter environment. By preliminarily processing the raw data collected in real time, the target detection process has timeliness, and the efficiency of the target detection process is improved through the preliminary processing of the raw data. By predicting the target appearance probability through the target detection model, the accuracy of target detection is improved, the phenomena of false detection and missed detection are reduced, and the reliability and accuracy of target detection are ensured. By using a fixed threshold to judge the prediction results of the target prediction data, the situation of false alarms is reduced. By aggregating the target point detection data, the error or noise interference in the target detection process is reduced, thereby improving the reliability and stability of the target detection results.

[0111] In several embodiments provided in the present application, it can be understood that each block in the flowchart or block diagram may represent a module, a program segment, or a part of code, and the module, the program segment, or the part of code includes one or more executable instructions for implementing the specified logical function. It should also be noted that in some alternative implementations, the functions marked in the blocks may occur in an order different from that marked in the drawings. For example, two consecutive blocks may actually be executed substantially in parallel, and they may sometimes be executed in the reverse order, depending on the functions involved.

[0112] If the described functions are implemented in the form of software function modules and sold or used as independent products, they can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present application, in essence, or the part that contributes to the prior art, or a part of this technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions for causing an electronic device to execute all or part of the steps of the methods described in the various embodiments of the present application. The aforementioned storage medium includes: USB flash drives, mobile hard disks, read-only memories (ROMs), random access memories (RAMs), magnetic disks, or optical discs, and other various media that can store program codes.

[0113] The specific embodiments described above further elaborate on the purpose, technical solution, and beneficial effects of the present application. It should be understood that the above description is only specific embodiments of the present application and is not used to limit the protection scope of the present application. It is particularly pointed out that for those skilled in the art, any modifications, equivalent replacements, improvements, etc. made within the spirit and principles of the present application shall be included in the protection scope of the present application.

Claims

1. A target detection method, characterized in that: A target detection model is pre-trained, and the target detection method includes: Collecting raw data in real time within a preset measurement range of the vehicle, and performing preliminary processing on the raw data to obtain scene data; Predicting the probability of target occurrence of the scene data based on the target detection model to obtain target prediction data; Performing a prediction result judgment on the target prediction data based on a fixed threshold value to obtain target point detection data according to the judgment result; And, all the target point detection data are condensed to obtain final target detection data.

2. The target detection method according to claim 1, characterized in that: The preliminary processing includes at least energy accumulation processing and pulse compression processing; wherein the energy accumulation processing includes: The raw data is formed into multi-dimensional matrix data; wherein the multi-dimensional matrix data at least includes a radar antenna channel dimension, a time dimension, and a pulse dimension; Slice the multidimensional matrix data based on the radar antenna channel dimension to obtain initial slice matrix data; The initial slice matrix data of a first preset value is acquired to acquire target slice matrix data based on the initial slice matrix data of the first preset value.

3. The target detection method according to claim 2, characterized in that: The pulse compression process comprises: Performing compression processing on the target slice matrix data based on the pulse dimension; The compressed target slice matrix data is subjected to Fourier transformation based on the time dimension to obtain the scene data.

4. The target detection method according to claim 3, characterized in that: The target detection model at least includes a feature extraction model, a target prediction model and a complex convolution model; the detecting of the scene data based on the target detection model includes: Extracting target semantic features from the scene data by using the feature extraction model; The target prediction model is used to predict the target occurrence probability of the target semantic features to obtain a target prediction matrix; The target prediction matrix of a second preset value is obtained through the complex convolution model to obtain the target prediction data based on the target prediction matrix of the second preset value.

5. The target detection method according to claim 4, characterized in that: The pre-training target detection model includes: detecting historical scene data through a pre-established target detection model to obtain historical target prediction data; Acquire historical target true value data to calculate the weight coefficient of the target detection model based on the historical target true value data and the historical target prediction data; The target detection model is optimized based on the weight coefficient to obtain a final target detection model.

6. The target detection method according to claim 5, characterized in that: The determining the target prediction data by fixing the threshold value includes: Based on the target prediction data, a preset processing algorithm is used to obtain a prediction value for each point in the target prediction data; Sorting the predicted values, and obtaining a fixed threshold value from the sorted predicted values ​​based on a preset false alarm rate; The predicted value is judged based on the fixed threshold value using a preset judgment rule to obtain target point detection data according to the judgment result.

7. The target detection method according to claim 6, characterized in that: The agglomeration process of all the target point detection data comprises: Acquire the target point detection data with the largest amplitude based on all the target point detection data; All target point detection data are condensed to the position of the target point detection data with the largest amplitude.

8. The target detection method according to claim 7, characterized in that: After obtaining the final target detection data, the method further includes: Acquire target detection data of a third preset value; Calculate and obtain the number of repetitions of the target detection data at the same coordinates based on the target detection data; It is determined whether the number of repetitions is greater than a preset number threshold. If so, it is determined that the target detection data exists. Otherwise, it is determined that the target detection data is abnormal and the abnormal target detection data is discarded.

9. A system based on the target detection method according to any one of claims 1 to 8, characterized in that: The system comprises: A data acquisition module, used to collect raw data in real time within a preset measurement range of the vehicle; A processing module, used for performing preliminary processing on the raw data to obtain scene data; A prediction module, used to predict the probability of target occurrence of the scene data based on the target detection model, and obtain target prediction data; A decision module, used to make a decision on the target prediction data by using a fixed threshold value, so as to obtain target point detection data according to the decision result; And, a condensation module is used to condense all the target point detection data to obtain final target detection data.

10. A computer-readable storage medium storing computer-executable instructions, characterized in that: When the computer executable instructions are executed by the control processor, the target detection method as described in any one of claims 1-8 is implemented.

Images