Target identification method, related device and terminal

Abstract

The invention discloses a target identification method, a related device and a terminal, which can be applied to automatic driving or auxiliary driving. The method comprises the steps of obtaining first data; the first data comprises at least one piece of information of a clustering target obtained based on millimeter wave radar measurement data; and processing the first data according to a scene division unit to obtain a scene classification result of the clustering target, wherein each scene processes the first data by a target classifier corresponding to the scene. The main contribution characteristics of the same target in different scenes are different, and scene classification is performed on the first data, so that the main contribution characteristics of the target in different scenes can be better identified, and the accuracy of target identification is improved.

Description

A target identification method, related device and terminal Technical Field

[0001] The present application relates to the field of autonomous driving technology, and in particular to a target recognition method, related devices, and terminals. Background Art

[0002] Millimeter-wave radar has the characteristics of wide bandwidth, short wavelength, strong penetration ability, high speed and distance measurement resolution, and is not easily affected by harsh environments such as darkness, strong light, clouds, fog, rain, snow, and hail. It has strong anti-interference ability and can work all day and all weather. It plays a vital role in solving the problem of road target recognition in autonomous driving and improving the safety of autonomous driving.

[0003] Autonomous driving road scenes are complex. In the process of target recognition and classification based on millimeter-wave radar, how to identify and classify targets more accurately is a problem that people have always hoped to solve.

[0004] Summary of the Invention

[0005] The embodiments of the present application provide a target recognition method, related devices and terminals, which can obtain more accurate target recognition results.

[0006] In a first aspect, an embodiment of the present application provides a target recognition method, the method comprising: acquiring first data; the first data comprising clustered target information obtained based on millimeter-wave radar measurement data; the information of the clustered target comprising one or more of sparse point cloud data, distance data, azimuth data, Doppler data, and occupancy grid map data (OGM data) of the clustered target; and processing the first data according to at least two scene classifiers to obtain a scene classification result of the clustered target. The first data may be single-category data or composite data comprising clustered target information, such as single-category data such as distance data, azimuth data, Doppler data, and occupancy grid map data (OGM data), or composite data consisting of, for example, distance data and Doppler data, or distance data and occupancy grid map data (OGM data), or composite data consisting of, for example, distance data, azimuth data, and Doppler data, or azimuth data, Doppler data, and occupancy grid map data (OGM data). The scene classifier for classifying the first data includes at least two different scenes. The scene classifier can process the first data and output its corresponding classification results. The main contributing features of the same target in different scenes are different. Scene classification of the first data according to different scene classifiers can better identify the main contributing features of the target in different scenes, thereby improving the accuracy of target recognition.

[0007] In a possible implementation of the first aspect, the scene classification results include: one or more of a radial target scene, a tangential target scene, a crossing target scene, a straight target scene, a stationary target scene, a field of view boundary target scene, an occluded target scene, a short-range target scene, a medium-range target scene, and a long-range target scene. The scene classifier includes at least two of the above scenes, and the scene classification results may be one or more. The first data may belong to only one of the above scenes, or may belong to multiple of the above scenes at the same time. The above scenes are distinguished and defined based on target features, and each scene is distinguished separately, and the target is identified by distinguishing the scenes, so that the target can be identified more accurately in each scene, thereby improving the accuracy of target recognition as a whole.

[0008] In a possible implementation of the first aspect, the radial target scene satisfies one or more of the following conditions: the radial velocity of the target is greater than or greater than or equal to a first threshold; the tangential velocity of the target is less than or less than or equal to a second threshold; and the ratio of the radial velocity of the target to the tangential velocity of the target is greater than or greater than or equal to a third threshold.

[0009] The tangential target scene satisfies one or more of the following conditions: the target tangential velocity is greater than or equal to a fourth threshold; the target radial velocity is less than or equal to a fifth threshold; and the ratio of the target tangential velocity to the radial velocity is greater than or equal to a sixth threshold.

[0010] The target-crossing scenario is a scenario in which the absolute value of the target's heading angle is greater than or equal to a seventh threshold and less than or equal to an eighth threshold.

[0011] The straight-ahead target scenario is a scenario in which the absolute value of the target's heading angle is greater than or equal to a ninth threshold, or the absolute value of the target's heading angle is less than or equal to a tenth threshold.

[0012] The stationary target scene satisfies one or more of the following conditions: a target radial velocity is less than or equal to an eleventh threshold and a tangential velocity is less than or equal to a twelfth threshold, and a target velocity is less than or equal to a thirteenth threshold.

[0013] The field of view boundary target scene is a scene in which the target is partially or completely located outside the field of view angle of the millimeter wave radar.

[0014] The occluded target scene is a scene in which there is a second target with an opening angle α2 to itself, and a first target with an opening angle α1 to itself, α2 and α1 have an overlapping angle β, and β is greater than or equal to the fourteenth threshold.

[0015] The short-range target scenario is a scenario where the distance between the target and the target itself is less than or equal to the fifteenth threshold.

[0016] The medium-distance target scene is a scene in which the distance between the target and the target itself is greater than or equal to the sixteenth threshold and less than or equal to the seventeenth threshold.

[0017] The long-distance target scenario is a scenario where the distance between the target and the self is greater than or equal to the eighteenth threshold, or a scenario where the distance between the target and the self threshold is greater than or equal to the eighteenth threshold and less than or equal to the nineteenth threshold.

[0018] By dividing and defining different scenarios, the first data can be classified into one or more scenarios for identification. Compared with directly identifying the first data in complex scenarios, the first data can be identified more accurately in specific scenarios, and the recognition accuracy in each scenario is improved, thereby improving the overall recognition accuracy.

[0019] A possible implementation of the first aspect includes obtaining second data based on the first data, the second data including classification feature parameters of the clustering target, the classification feature parameters including one or more of the speed, circumference, length, width, height, heading angle, distance, and opening angle of the clustering target, and the classification feature parameters may also include other parameters of the clustering target, which is not limited in this embodiment of the present application; the first data is processed according to the scene classifier and the second data to obtain a scene classification result of the clustering target.

[0020] The second data is obtained after processing the first data and is used to classify the first data for different scenarios. The second data can be single-category data such as the target's speed, circumference, length, width, height, heading angle, distance, or angle of attack; composite data including multiple types of the aforementioned data; or composite parameters derived from calculations of the aforementioned data that can serve as a basis for classification. The second data obtained by processing the first data has more distinct classification feature parameters, can support scene classification of the first data, and obtain scene classification results for clustered targets, thereby more accurately identifying the first data.

[0021] In a possible implementation of the first aspect, the first data includes a first cluster target, and the scene classification result of the first cluster target includes that the first cluster target belongs to a first scene; a first target classifier corresponds to the first scene, and the first target classifier classifies the first data. After the first data is divided into different scenes, each scene has a corresponding target classifier to classify the first data. The target classifier can be optimized and trained for its corresponding scene so that the classifier is suitable for target classification in the scene. Different target classifiers can better identify the main contributing features of the target in a separate scene, thereby obtaining a more accurate classification result.

[0022] Optionally, the first clustered objects may belong to multiple scenes simultaneously, for example, both the first and second scenes. In this case, the first data is fed into the first and second object classifiers, respectively, which then output classification results. Similarly, the first clustered objects may belong to three or more scenes. The identification method is the same as described above and will not be further described here.

[0023] In a possible implementation of the first aspect, the first target classifier is a traditional machine learning classifier, a deep learning classifier or a multi-frame fusion classifier, and the traditional machine learning classifier includes: a naive Bayes classifier, a logistic regression classifier, a support vector machine classifier (SVM classifier), a random forest classifier (RF classifier), a decision tree classifier, an adaptive boosting classifier (AdaBoost classifier), and an extreme gradient boosting classifier (XGBoost classifier); the deep learning classifier includes: a fully connected neural network classifier, a convolutional neural network classifier, and a deep residual neural network classifier, and the multi-frame fusion classifier includes a multi-frame fusion classifier that uses Kalman filtering, hidden Markov, and naive Bayes technology to perform multi-frame joint processing based on the above-mentioned traditional machine learning classifier or deep learning classifier. For different scenarios, the classifier type may be the same or different to better adapt to the corresponding scenario and improve the accuracy and efficiency of recognition.

[0024] In a possible implementation of the first aspect, the scene classification result of the first cluster target includes that the first cluster target belongs to at least two scenes, and the first cluster target is classified based on classifiers corresponding to the at least two scenes;

[0025] If the same target has target classification results output in multiple classifiers, the classification result of the target is determined after performing a first operation on the multiple target classification results of the same target. When the target is in multiple scenes at the same time, there may be differences in the classification results of multiple classifiers. The accuracy of the classification results can be improved by performing operations on the classification results of different classifiers and then outputting them. Optionally, the first operation is to perform a weighted average of the output results of multiple target classifiers based on their reliability, and the final result is the final classification result of the target; for example, the classification result of the target classifier with the highest reliability is selected as the final classification result of the target. The comprehensive utilization of the classification results of multiple target classifiers can improve the accuracy and credibility of the final target recognition result.

[0026] In one possible implementation of the first aspect, the scene classification result of the first clustered target includes that the first clustered target belongs to at least two scenes, and the first clustered target is classified based on the target classifiers corresponding to the at least two scenes; if the same target is only outputted by a single classifier, the classification result of the target by that classifier is the final target classification result. If the target is only outputted by one classifier, that result is outputted as the final result, thereby improving recognition efficiency.

[0027] In a second aspect, the present application provides a target recognition device, comprising an acquisition module and a processing module.

[0028] The acquisition module is used to acquire first data, wherein the first data includes cluster target information obtained based on millimeter-wave radar measurement data; the cluster target information includes one or more of sparse point cloud data, distance data, azimuth data, Doppler data, and occupancy grid map data (OGM data) of the cluster target; and the processing module is used to process the first data to obtain a scene classification result for the cluster target. The processing module can classify at least two scenes, and may also classify three or more scenes, but this does not mean that the classification results of the first data must be two or more. The processing module has the ability to classify two or more scenes, but the first data may belong to one or more scenes. The first data can be single-category data including cluster target information or multidimensional composite data, such as single-category data such as distance data and azimuth data, composite data such as distance data and Doppler data, or composite data such as distance data, azimuth data, and Doppler data.

[0029] In a possible implementation of the second aspect, the scene classification result includes:

[0030] One or more of radial target scene, tangential target scene, traversing target scene, straight target scene, stationary target scene, field of view boundary target scene, occluded target scene, short-range target scene, medium-range target scene, and long-range target scene.

[0031] In a possible implementation of the second aspect, the radial target scene satisfies one or more of the following conditions: the radial velocity of the target is greater than or greater than or equal to a first threshold; the tangential velocity of the target is less than or less than or equal to a second threshold; and the ratio of the radial velocity of the target to the tangential velocity of the target is greater than or greater than or equal to a third threshold.

[0032] The tangential target scene satisfies one or more of the following conditions: the target tangential velocity is greater than or equal to a fourth threshold; the target radial velocity is less than or equal to a fifth threshold; and the ratio of the target tangential velocity to the radial velocity is greater than or equal to a sixth threshold.

[0033] The target-crossing scenario is a scenario in which the absolute value of the target's heading angle is greater than or equal to a seventh threshold and less than or equal to an eighth threshold.

[0034] The straight-ahead target scenario is a scenario in which the absolute value of the target's heading angle is greater than or equal to a ninth threshold, or the absolute value of the target's heading angle is less than or equal to a tenth threshold.

[0035] The stationary target scene satisfies one or more of the following conditions: a target radial velocity is less than or equal to an eleventh threshold and a tangential velocity is less than or equal to a twelfth threshold, and a target velocity is less than or equal to a thirteenth threshold.

[0036] The field of view boundary target scene is a scene in which the target is partially or completely located outside the field of view angle of the millimeter wave radar.

[0037] The occluded target scene is a scene in which there is a second target with an opening angle α2 to itself, and a first target with an opening angle α1 to itself, α2 and α1 have an overlapping angle β, and β is greater than or equal to the fourteenth threshold.

[0038] The short-range target scenario is a scenario where the distance between the target and the target itself is less than or equal to the fifteenth threshold.

[0039] The medium-distance target scene is a scene in which the distance between the target and the target itself is greater than or equal to the sixteenth threshold and less than or equal to the seventeenth threshold.

[0040] The long-distance target scenario is a scenario where the distance between the target and the self is greater than or equal to the eighteenth threshold, or a scenario where the distance between the target and the self threshold is greater than or equal to the eighteenth threshold and less than or equal to the nineteenth threshold.

[0041] A possible implementation of the second aspect includes obtaining second data based on the first data, the second data including classification feature parameters of the clustering target, the classification feature parameters including one or more of the speed, circumference, length, width, height, heading angle, distance, and opening angle of the clustering target, and the classification feature parameters may also be other feature parameters of the clustering target, which is not limited in this application; the first data is processed according to the scene classifier and the second data to obtain a scene classification result of the clustering target.

[0042] The second data is obtained after processing the first data and is used to classify the first data into scenes. The second data can be a single type of data including the target's speed, circumference, length, width, height, heading angle, distance, and opening angle, or it can be composite data including multiple types of the above data, or it can be a composite parameter obtained after calculating the above multiple data and can be used as a basis for classification.

[0043] In a possible implementation of the second aspect, the detection device also includes a first target classifier, the first data includes a first cluster target, and the scene classification result of the first cluster target includes that the first cluster target belongs to a first scene; the first target classifier corresponds to the first scene, and the first target classifier is used to classify the first data.

[0044] Optionally, the first target classifier is a traditional machine learning classifier, a deep learning classifier or a multi-frame fusion classifier, and the traditional machine learning classifier includes: naive Bayes classifier, logistic regression classifier, support vector machine classifier (SVM classifier), random forest classifier (RF classifier), decision tree classifier, adaptive boosting classifier (AdaBoost classifier), extreme gradient boosting classifier (XGBoost classifier); the deep learning classifier includes: fully connected neural network classifier, convolutional neural network classifier, deep residual neural network classifier, and the multi-frame fusion classifier includes a multi-frame fusion classifier based on the above-mentioned traditional machine learning classifier or deep learning classifier, which uses Kalman filtering, hidden Markov, and naive Bayes technology to perform multi-frame joint processing. For different scenarios, the classifier type may be the same or different to better adapt to the corresponding scenario and improve the accuracy and efficiency of recognition.

[0045] In a possible implementation of the second aspect, the scene classification result of the first cluster target includes that the first cluster target belongs to at least two scenes, and the first cluster target is classified based on the target classifiers corresponding to the at least two scenes; if the same target has target classification results output in multiple target classifiers, then after performing a first operation on the multiple target classification results of the same target, the classification result of the target is determined. Optionally, the first operation is to perform a weighted average of the output results of multiple target classifiers according to their reliability, and the final result is the final classification result of the target; for another example, the classification result of the target classifier with the highest reliability is selected as the final classification result of the target. Comprehensively utilizing the classification results of multiple target classifiers can improve the accuracy and credibility of the final target recognition result.

[0046] In a possible implementation of the second aspect, the scene classification result of the first clustering target includes that the first clustering target belongs to at least two scenes, and the first clustering target is classified based on the target classifier corresponding to the at least two scenes; if the same target has a target classification result output only in a single target classifier, then the classification result of the target by the target classifier is the final target classification result.

[0047] In a third aspect, the present application provides a terminal, the terminal comprising the target recognition device according to any possible implementation of the second aspect. Optionally, the terminal is a vehicle, a drone, or a robot.

[0048] In a fourth aspect, the present application provides a computer-readable storage medium, wherein the computer-readable storage medium is used to store instructions, and when the instructions are executed, any one of the identification methods in the first aspect is implemented.

[0049] Regarding the implementation effects brought about by the implementation methods corresponding to the second, third and fourth aspects, please refer to the introduction of the above various implementation methods, which will not be repeated here. BRIEF DESCRIPTION OF THE DRAWINGS

[0050] FIG1 is a schematic diagram of data trends of an identification method provided by an embodiment of the present application;

[0051] FIG2 is a schematic diagram of a data processing method of a classifier provided in an embodiment of the present application;

[0052] FIG3 is a schematic diagram of a radial target scene provided by an embodiment of the present application;

[0053] FIG4 is a schematic diagram of a tangential target scenario provided by an embodiment of the present application;

[0054] FIG5 is a schematic diagram of a target crossing scenario provided by an embodiment of the present application;

[0055] FIG6 is a schematic diagram of a straight-line target scenario provided by an embodiment of the present application;

[0056] FIG7 is a schematic diagram of a field of view boundary target scene provided by an embodiment of the present application;

[0057] FIG8 is a schematic diagram of a target occlusion scenario provided by an embodiment of the present application;

[0058] FIG9 is a schematic diagram of a target scene at different distances provided by an embodiment of the present application;

[0059] FIG10 is a flow chart of a target recognition method provided in an embodiment of the present application;

[0060] FIG11 is a schematic diagram of a target recognition device provided in an embodiment of the present application;

[0061] FIG12 is a schematic diagram of another target recognition device provided in an embodiment of the present application. DETAILED DESCRIPTION

[0062] In order to make the purpose, technical solutions and advantages of the embodiments of the present application clearer, the embodiments of the present application will be further described in detail below with reference to the accompanying drawings.

[0063] For ease of understanding, the following examples provide some explanations of concepts related to the embodiments of the present application for reference.

[0064] 1. Millimeter-wave radar

[0065] Millimeter-wave radar operates in the millimeter-wave band. Millimeter waves typically refer to electromagnetic waves in the 20-300 GHz frequency range (wavelengths of 1-10 mm). Millimeter-wave radars are characterized by their small size, light weight, and high spatial resolution. They are also highly resilient to traditional smoke, fog, and dust, and can operate in all weather conditions and around the clock.

[0066] 2. Clustering Target

[0067] The data obtained by millimeter-wave radar measurement is clustered and analyzed to divide the data into data clusters consisting of similar objects, called cluster targets. The objects in a cluster are similar to other objects in the same cluster, but different from the data in other clusters.

[0068] 3. Point cloud data

[0069] Point cloud data refers to a collection of vectors in a multidimensional coordinate system. The data measured by millimeter-wave radar can be different points in the system, and the collection of these points constitutes point cloud data.

[0070] 4. Doppler data

[0071] The Doppler effect refers to the difference between the frequency of the vibration received by an observer and the frequency emitted by the source when a vibration source such as sound, light, or radio waves moves at a relative velocity V. Doppler data, measured by millimeter-wave radar using the Doppler effect, reflects the relative velocity of the target and the observer.

[0072] 5. Azimuth data

[0073] Azimuth data refers to data reflecting the relative azimuth relationship between the target observed by the millimeter wave radar and itself.

[0074] 6. Occupancy grid map

[0075] Also known as the Occupancy Grid Map, the millimeter-wave radar's observation environment is divided into grids and each grid is filled with a binary value, where 0 indicates the grid is occupied and 1 indicates it is not occupied. Alternatively, each grid is stored with a probability value, where the larger the value, the greater the likelihood that the grid is dedicated.

[0076] 7. Heading angle

[0077] In the ground coordinate system, the angle between the target center of mass velocity and the x-axis is the heading angle.

[0078] 8. Processing module

[0079] The processing module is a processing unit for processing and identifying the first data.

[0080] The processing unit may include one or more processors. It should be understood that in the embodiments of the present application, for the convenience of explaining the computing function, it is described as a processor. In the specific implementation process, the processor may include a device with a computing function. For example, at least one processor may include one or more of the following devices: a central processing unit (CPU), an application processor (AP), a time-to-digital converter (TDC), a filter, a graphics processing unit (GPU), a microprocessor (MPU), an application specific integrated circuit (ASIC), an image signal processor (ISP), a digital signal processor (DSP), a field programmable gate array (FPGA), a complex programmable logic device (CPLD), a coprocessor (to assist the central processing unit in completing the corresponding processing and application), a microcontroller unit (MCU), and / or a neural-network processing unit (NPU), etc.

[0081] Optionally, the processing unit may be located inside or outside the target recognition device.

[0082] Alternatively, in some designs, the processing module may include multiple components, some of which are located inside the device and some of which are located outside the device. For example, the processing module may include a digital-to-analog conversion module, a filtering module, a time-of-flight (TOF) decoding module, and a point cloud generation module, wherein the digital-to-analog conversion module, the filtering module, and the TOF decoding module are located inside the device, while the point cloud generation module is located outside the device.

[0083] 9. Classifier

[0084] The functional unit for classifying or identifying data may be composed of a processing module, or may refer to a virtual software algorithm functional unit in the processing module that has corresponding processing functions.

[0085] 10. Classification feature parameters

[0086] The speed, shape parameters, length, width, height, circumference, heading angle, distance, angle of flare of the target observed by the millimeter wave radar, or a combination or deformation of the above data, or other data that can be used to distinguish the scene to which the first data belongs, are not limited in this application and are collectively referred to as classification feature parameters.

[0087] 11. Main Contributing Features

[0088] In a specific scenario, the combination of feature data that provides the main basis for target recognition is called the main contributing feature. For example, in a radial target scenario, the target's speed is the main contributing feature with a weight of 50%, and the target's heading angle is the main contributing feature with a weight of 20%; in a straight target scenario, the target's speed is the main contributing feature with a weight of 30%, and the target's heading angle is the main contributing feature with a weight of 40%; for another example, in a stationary target scenario, the target's length is the main contributing feature with a weight of 10%, and in an occluded target scenario, it is the main contributing feature with a weight of 10%. The composition of the main contributing features and the weight ratio of the features in different scenarios may be the same or different, and this application does not impose any restrictions on this.

[0089] In the embodiments of the present application, for the number of nouns, unless otherwise specified, it means "singular noun or plural noun", that is, "one or more". "At least one" means one or more, and "plural" means two or more. "And / or" describes the association relationship of associated objects, indicating that there may be three relationships. For example, A and / or B can mean: A exists alone, A and B exist at the same time, and B exists alone, where A and B can be singular or plural. Unless otherwise specified, the character " / " generally indicates that the previous and next associated objects are in an "or" relationship. For example, A / B means: A or B. "At least one of the following items" or similar expressions refers to any combination of these items, including any combination of single items or plural items. For example, at least one of a, b, or c means: a, b, c, a and b, a and c, b and c, or a and b and c, where a, b, c can be single or multiple.

[0090] In the embodiments of this application, ordinal numbers such as "first" and "second" are used to distinguish multiple objects and are not used to limit the size, content, order, timing, application scenario, priority, or importance of the multiple objects. For example, the first processing module and the second processing module can be the same processing module or different processing modules, and such names do not indicate differences in the structure, location, priority, application scenario, or importance of the two processing modules.

[0091] In the embodiments of the present application, "connection" can be a direct connection or an indirect connection; in addition, it can refer to an electrical connection or a communication connection; for example, the connection between two electrical components A and B can refer to a direct connection between A and B, or it can refer to an indirect connection between A and B through other electrical components or a connection medium, so that electrical signals can be transmitted between A and B; for another example, the connection between two devices A and B can refer to a direct connection between A and B, or it can refer to an indirect connection between A and B through other communication devices or communication media, as long as communication between A and B can be carried out.

[0092] In the various embodiments of the present application, unless otherwise specified or there is a logical conflict, the terms and / or descriptions between the various embodiments are consistent and can be referenced by each other. The technical features in different embodiments can be combined to form new embodiments according to their inherent logical relationships.

[0093] Those skilled in the art will understand that all or part of the steps to implement the above embodiments may be accomplished by hardware, or by a program to instruct the relevant hardware, and the program may be stored in a computer-readable storage medium, which may be a read-only memory, a disk, or an optical disk, etc.

[0094] The above introduces some concepts involved in the embodiments of the present application. The following introduces the technical features of the embodiments of the present application.

[0095] Millimeter-wave radar is widely used in the field of autonomous driving. In millimeter-wave radar target recognition, a single classifier is widely used to identify road targets in complex environments. However, the same target will exhibit different main contributing features in different scenarios. Using a single classifier to solve this problem will compromise the main contributing features of the target in multiple scenarios, resulting in loss or degradation of recognition performance.

[0096] To this end, the present application provides a target recognition method. Please refer to Figure 1, which is a data flow diagram of a recognition method provided by the present application, including: obtaining first data; the first data includes cluster target information obtained based on millimeter wave radar measurement data; the information of the cluster target includes one or more of sparse point cloud data, distance data, azimuth data, Doppler data, and occupancy grid map data (OGM data) of the cluster target; processing the first data according to at least two scene classifiers to obtain a scene classification result of the cluster target. The method is partially executed by the scene division unit in Figure 1. The scene division unit can be integrated into the hardware or software of the millimeter wave radar, or it can be a hardware or software functional unit that can implement this function in a mobile data center (MDC) or other processing module with data processing function. The target recognition method is applied to hardware units such as millimeter wave radar, mobile data center (MDC), advanced driver assistant system (ADAS) and vehicle control unit (VCU).

[0097] Please refer to Figure 1. The first data is classified in a scene classification unit composed of two or more scene classifiers. The classification result is one or more scenes, and then enters the target classifier of the corresponding scene, and finally outputs the classification result.

[0098] In a possible implementation, the scene classification results include one or more of: radial target scenes, tangential target scenes, cross-target scenes, straight-ahead target scenes, stationary target scenes, field-of-view boundary target scenes, occluded target scenes, short-range target scenes, medium-range target scenes, and long-range target scenes. The scene classifier includes at least two of the above scenes, and the scene classification results can be one or more. The first data may belong to only one of the above scenes, or may belong to multiple of the above scenes at the same time. The above scenes are common scenes in target recognition. Distinguishing each scene individually and identifying targets based on the scenes can improve the accuracy of target recognition.

[0099] Optionally, the data may pass through the classifier in a serial or parallel manner, or a combination of serial and parallel manners, as shown in FIG2 , where the classifier may be a scene classifier or an object classifier.

[0100] Specifically, a classifier is used to determine a classification result of a scene or target. Each scene is further introduced below. It should be noted that, in addition to the scenes introduced below, the multiple classifiers can also include other classifiers with scene classification functions or target classification functions, and the embodiments of the present application do not limit the upper limit of the number of classifiers included. In addition, two or more partial classifiers among these multiple classifiers can also be combined to form a classifier with more comprehensive functions, or a classifier can also be split into multiple classifiers with more detailed and specific judgment rules. The splitting and combination of classifiers are not used to limit the scope of the embodiments of the present application.

[0101] A possible implementation method, please refer to Figure 3, which is a schematic diagram of a radial target scenario. The radial target scenario satisfies one or more of the following conditions: the radial velocity of the target is greater than or equal to a first threshold; the tangential velocity of the target is less than or equal to a second threshold; the ratio of the radial velocity of the target to the tangential velocity of the target is greater than or equal to a third threshold. The radial velocity is the component of the target velocity on the straight line connecting the detector and the target, and the tangential velocity is the component of the target velocity in the vertical direction of the straight line connecting the detector and the target. Due to the working principle of millimeter-wave radar, its radial velocity measurement is more accurate than the tangential velocity measurement. Therefore, defining a radial target scenario can better utilize the performance advantages of millimeter-wave radar and improve recognition accuracy and efficiency.

[0102] For a possible implementation, please refer to Figure 4, which is a schematic diagram of a tangential target scenario, wherein the tangential target scenario satisfies one or more of the following conditions: the target tangential velocity is greater than or equal to a fourth threshold; the target radial velocity is less than or equal to a fifth threshold; and the ratio of the target tangential velocity to the radial velocity is greater than or equal to a sixth threshold.

[0103] The radial velocity is the component of the target velocity on the line connecting the probe and the target, and the tangential velocity is the component of the target velocity in the direction perpendicular to the line connecting the probe and the target.

[0104] For a possible implementation, please refer to FIG5 , which is a schematic diagram of a target crossing scenario, wherein the target crossing scenario is a scenario in which the absolute value of the target heading angle is greater than or equal to the seventh threshold and less than or equal to the eighth threshold.

[0105] The heading angle is the angle between the target's center of mass and the x-axis in the ground coordinate system. In this scenario, the target's motion is generally perpendicular to the x-axis, which is typically aligned with the target's path. This is known as a "crossing target" scenario.

[0106] As a possible implementation, please refer to Figure 6, which is a schematic diagram of a straight-moving target scenario, where the absolute value of the target's heading angle is greater than or equal to the ninth threshold, or the absolute value of the target's heading angle is less than or equal to the tenth threshold. In this scenario, the target's motion trend is generally along the x-axis.

[0107] In a possible implementation manner, the stationary target scene satisfies one or more of the following conditions: the target radial velocity is less than or equal to an eleventh threshold and the tangential velocity is less than or equal to a twelfth threshold, and the target velocity is less than or equal to a thirteenth threshold.

[0108] When the target speed is lower than a certain threshold, it can be treated as a stationary target.

[0109] As a possible implementation, please refer to FIG7 , which is a schematic diagram of a field of view boundary target scene. The field of view boundary target scene is a scene in which the target is partially or completely located outside the field of view of the millimeter wave radar.

[0110] Due to the characteristics of millimeter-wave radar, targets outside its field of view will be deformed, making their features different from those of targets within the field of view. When the target is partially or completely outside the field of view of the millimeter-wave radar, processing such target information with a separate scene and classifier can improve the accuracy and efficiency of target recognition.

[0111] A possible implementation method, please refer to Figure 8, which is a schematic diagram of an occluded target scene. The occluded target scene is a scene in which there is a second target with an angle α2 to itself, and there is a first target with an angle α1 to itself. α2 and α1 have an overlapping angle β, and β is greater than or equal to the fourteenth threshold.

[0112] The range corresponding to the angle β is the occluded area of ​​target B, and the remaining area is the visible area. When the target is occluded, the characteristics of the remaining visible area may change. Therefore, this scenario is distinguished separately to improve the accuracy of target recognition in this scenario.

[0113] For a possible implementation, please refer to FIG9 , which is a schematic diagram of target scenes at different distances. According to the distance between the target and the millimeter-wave radar, the scenes are divided into short-range target scenes (Short Range), medium-range target scenes (Medium Range) and long-range target scenes (Long Range). The short-range target scene (Short Range) is a scene where the distance between the target and itself is less than or equal to the fifteenth threshold; the medium-range target scene (Medium Range) is a scene where the distance between the target and itself is greater than or equal to the sixteenth threshold, and less than or equal to the seventeenth threshold; the long-range target scene (Long Range) is a scene where the distance between the target and itself is greater than or equal to the eighteenth threshold, or a scene where the distance between the target and itself is greater than or equal to the eighteenth threshold, and less than or equal to the nineteenth threshold. As shown in FIG7 , three different scenes are defined according to the distance between the target and the millimeter-wave radar. The echo characteristics and point cloud characteristics generated by the target are different at different distances. Distinguishing target scenes at different distances is conducive to improving the recognition accuracy of targets at different distances.

[0114] A possible implementation method includes: obtaining second data based on the first data, the second data including classification feature parameters of the clustering target, the classification feature parameters including one or more of the speed, circumference, length, width, height, heading angle, distance, and opening angle of the clustering target; processing the first data according to a scene classifier and the second data to obtain a scene classification result of the clustering target.

[0115] The second data is obtained after processing the first data and is used to classify the first data into scenes. The second data can be a single type of data including the target's speed, circumference, length, width, height, heading angle, distance, angle of attack, etc., or it can be composite data including multiple types of the above data, or it can be a composite parameter obtained after calculating the above multiple data and can be used as a basis for classification.

[0116] A possible implementation includes: the first data includes a first cluster target, and the scene classification result of the first cluster target includes that the first cluster target belongs to a first scene; a first classifier corresponds to the first scene, and the first classifier classifies the first data. Please refer to Figure 10, which is a flowchart of a target recognition method provided by this application, S101, obtaining first data; optionally, executing step S102 to obtain second data based on the first data; S103, performing scene classification on the first data to obtain a scene classification result, which is performed by a scene division unit; step S104, based on the result of the scene classification, the first data is classified by the target classifier corresponding to the scene, which is performed by the target classification; finally, step S105, outputting the target classification result. For the scene classification result obtained in step S103, please refer to the corresponding relationship between the scene classification result and the target classifier in Figure 1. After obtaining the scene classification result, the first data is transmitted to the target classifier corresponding to the scene for processing. For example, the first target classifier corresponds to the first scene. When the scene classification result of the first data contains the first scene, the first data is processed by the first target classifier.

[0117] After the first data is divided into different scenes, it is sent to the classifier corresponding to each scene for recognition and processing. The classifier corresponding to each scene outputs the recognition result for that scene. Because the classifier corresponding to each scene only processes target recognition in that scene, it can be trained and adjusted specifically for that scene and classifier, achieving better performance in that scene, thereby improving target recognition accuracy.

[0118] Optionally, the classifier is a traditional machine learning classifier, a deep learning classifier or a multi-frame fusion classifier, and the traditional machine learning classifier includes: naive Bayes classifier, logistic regression classifier, support vector machine classifier (SVM classifier), random forest classifier (RF classifier), decision tree classifier, adaptive boosting classifier (AdaBoost classifier), extreme gradient boosting classifier (XGBoost classifier); the deep learning classifier includes: fully connected neural network classifier, convolutional neural network classifier, deep residual neural network classifier; the multi-frame fusion classifier includes a multi-frame fusion classifier based on the above-mentioned traditional machine learning classifier or deep learning classifier using Kalman filtering, hidden Markov, and naive Bayes technology for multi-frame joint processing. The classifiers corresponding to different scenarios can be classifiers of the same type or different types, and this application does not impose any restrictions on this.

[0119] A possible implementation method includes that the scene classification result of the first clustering target includes that the first clustering target belongs to at least two scenes, and the first clustering target is classified based on the classifiers corresponding to the at least two scenes; if the same target has target classification results output in multiple classifiers, then after performing a first operation on the multiple target classification results of the same target, the classification result of the target is determined; if the same target has a target classification result output only in a single classifier, then the classification result of the target by the classifier is the final target classification result. For example, the target is an occluded pedestrian in a close-range scene. The target belongs to both the close-range target scene and the occluded target scene. Therefore, the target is identified by the target classifiers corresponding to the two scenes. Due to the different target recognition capabilities of the target classifiers in the scenes, for example, the occluded target scene is more sensitive to the characteristics of pedestrians, while the classifier corresponding to the close-range target scene is more sensitive to vehicle targets, it is possible that only the target classifier corresponding to the occluded target scene outputs a recognition result of a pedestrian, and the target classifier corresponding to the close-range target scene has no output recognition result, then the classification result output for the target is a pedestrian; for another example, the target classifier corresponding to the occluded target scene outputs a recognition result of a pedestrian, and the target classifier corresponding to the close-range target scene outputs a recognition result of a tree. After the first calculation, the classification result output for the target is a pedestrian.

[0120] If the same target has target classification results output in multiple classifiers, the classification result of the target is determined after the first operation is performed on the multiple target classification results of the same target. When the target is in multiple scenes at the same time, there may be differences in the classification results of multiple classifiers. The accuracy of the classification results can be improved by performing operations on the classification results of different classifiers and then outputting them. Optionally, the first operation is to perform a weighted average of the output results of multiple target classifiers based on their reliability, and the final result is the final classification result of the target; for example, the classification result of the target classifier with the highest reliability is selected as the final classification result of the target. The comprehensive utilization of the classification results of multiple target classifiers can improve the accuracy and credibility of the final target recognition result. When the target has a result output in only one classifier, outputting that result as the final result can improve recognition efficiency.

[0121] As a result of scene segmentation of the first data, the first clustered target may belong to two or more scenes. For example, if the first clustered target belongs to both the radial target scene and the short-range target scene, classification results for the target are output for both scenes. After further processing of the two results, the final classification result for the target is obtained, completing the recognition of the target. For example, if the clustered target belongs to both the occluded target scene and the straight-ahead target scene, but only the classifier corresponding to the straight-ahead target scene has a classification result for the target, then this classification result is the final recognition result for the target.

[0122] In one possible implementation, the scenes may be divided into two or more, for example, the first scene corresponds to a short-range target scene and a medium-range target scene, and the second scene corresponds to a long-range target scene; optionally, the first scene corresponds to a short-range target scene, and the second scene corresponds to a medium-range target scene and a long-range target scene; optionally, the first scene corresponds to a radial target scene, and the second scene corresponds to a tangential target scene; optionally, the first scene corresponds to a transverse target scene, and the second scene corresponds to a straight-moving target scene; optionally, the first scene corresponds to a stationary target scene, and the second scene corresponds to a non-stationary target scene; optionally, the first scene corresponds to a field of view boundary target scene, and the second scene corresponds to a non-field of view boundary target scene; optionally, the first scene corresponds to an occluded target scene, and the second scene corresponds to a non-occluded target scene. Other scene combinations are possible and will not be elaborated here.

[0123] Optionally, when divided into three scenes, for example, the first scene corresponds to a stationary target scene, the second scene corresponds to a radial target scene outside the stationary target scene, and the third scene corresponds to a tangential target scene outside the stationary target scene; Optionally, the first scene corresponds to a stationary target scene, the second scene corresponds to a crossing target scene outside the stationary target scene, and the third scene corresponds to a straight target scene outside the stationary target scene; Optionally, the first scene corresponds to a short-range target scene, the second scene corresponds to a medium-range target scene, and the third scene corresponds to a long-range target scene; Optionally, the first scene corresponds to a field of view boundary target scene, and the second scene corresponds to a radial target scene outside the field of view boundary target scene The third scene corresponds to a tangential target scene outside the field of view boundary target scene; optionally, the first scene corresponds to the field of view boundary target scene, the second scene corresponds to a cross-target scene outside the field of view boundary target scene, and the third scene corresponds to a straight-ahead target scene outside the field of view boundary target scene; optionally, the first scene corresponds to an occlusion target scene, the second scene corresponds to a radial target scene outside the occlusion target scene, and the third scene corresponds to a tangential target scene outside the occlusion target scene; optionally, the first scene corresponds to an occlusion target scene, the second scene corresponds to a cross-target scene outside the occlusion target scene, and the third scene corresponds to a straight-ahead target scene outside the occlusion target scene. There may be other scene combinations, which will not be elaborated here.

[0124] Optionally, the scenes can be further divided into sub-scenes. For example, the short-range, medium-range, and long-range scenes in the stationary target scene constitute three scenes; for another example, the field of view boundary scene is divided into radial target scenes and tangential target scenes, constituting two scenes; for another example, the occluded target scene and the non-occluded target scene in the radial target constitute two scenes. The above scenes can be combined to form the required scene division. For example, the short-range, medium-range, and long-range scenes in the stationary target scene and the radial target scene and the tangential target scene constitute a scene division containing five scenes; for another example, the occluded target scene and the non-occluded target scene in the radial target and the short-range, medium-range, and long-range scenes in the stationary target scene, as well as the straight-ahead target scene constitute a scene division containing six scenes. The above scenes can also be divided and combined into more scenes according to the needs of target recognition, which will not be repeated here. This application does not limit the number of scene divisions.

[0125] An embodiment of the present application provides a target recognition device. Please refer to FIG11 , which is a schematic diagram of the recognition device provided by the present application. The device includes an acquisition module and a processing module. The acquisition module is configured to acquire first data, wherein the first data includes clustered target information obtained based on millimeter-wave radar measurement data. The clustered target information includes one or more of sparse point cloud data, distance data, azimuth data, Doppler data, and occupancy grid map data (OGM data) of the clustered target. The processing module is configured to process the first data to obtain a scene classification result for the clustered target. The processing module is capable of classifying at least two scenes, and may also classify three or more scenes. However, this does not necessarily mean that the classification results of the first data must be two or more. The processing module has the ability to classify two or more scenes, but the first data may belong to one or more scenes. The first data can be single-category data or composite data including clustered target information, such as single-category data such as distance data and azimuth data, composite data consisting of distance data and Doppler data, or composite data consisting of distance data, azimuth data, and Doppler data. Optionally, the target recognition device described in the embodiment of the present application may be a millimeter wave radar, or it may be a hardware unit such as a mobile data center (MDC), an advanced driver assistant system (ADAS), a vehicle control unit (VCU), an electronic control unit (ECU), or a system on chip (SoC).

[0126] An embodiment of the present application provides a target identification device. Please refer to Figure 12, which is a schematic diagram of an identification device provided by the present application. The device includes at least one processor 1201 and an interface circuit 1202. The processor is connected to the interface circuit. The interface circuit is used to receive data from other devices or send signals from the processor to other devices. The processor implements the above-mentioned target identification method provided by the embodiment of the present application through logic circuits or executing code instructions.

[0127] It should be understood that the processors mentioned in the embodiments of the present application can be implemented by hardware or software. When implemented by hardware, the processor can be a logic circuit, an integrated circuit, etc. When implemented by software, the processor can be a general-purpose processor that is implemented by reading software code stored in a memory.

[0128] Exemplarily, the processor may be a central processing unit (CPU), other general-purpose processors, digital signal processors (DSP), application-specific integrated circuits (ASIC), field programmable gate arrays (FPGA), or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. The general-purpose processor may be a microprocessor or any conventional processor, etc.

[0129] It should be understood that the memory mentioned in the embodiments of the present application may be a volatile memory or a non-volatile memory, or may include both volatile and non-volatile memories. Among them, the non-volatile memory may be a read-only memory (ROM), a programmable read-only memory (PROM), an erasable programmable read-only memory (EPROM), an electrically erasable programmable read-only memory (EEPROM), or a flash memory. The volatile memory may be a random access memory (RAM), which is used as an external cache. By way of example and not limitation, many forms of RAM are available, such as static random access memory (SRAM), dynamic random access memory (DRAM), synchronous dynamic random access memory (SDRAM), double data rate synchronous dynamic random access memory (DDR SDRAM), enhanced synchronous dynamic random access memory (ESDRAM), synchronous link dynamic random access memory (SLDRAM), and direct RAM bus random access memory (DR RAM).

[0130] It should be noted that when the processor is a general-purpose processor, DSP, ASIC, FPGA or other programmable logic device, discrete gate or transistor logic device, discrete hardware component, the memory (storage module) can be integrated into the processor.

[0131] It should be noted that the memory described herein is intended to include, but not be limited to, these and any other suitable types of memory.

[0132] In a possible implementation, the scene classification results include one or more of: radial target scenes, tangential target scenes, cross-target scenes, straight-ahead target scenes, stationary target scenes, field-of-view boundary target scenes, occluded target scenes, short-range target scenes, medium-range target scenes, and long-range target scenes. The scene classifier includes at least two of the above scenes, and the scene classification results can be one or more. The first data may belong to only one of the above scenes, or may belong to multiple of the above scenes at the same time. The above scenes are common scenes in target recognition. Distinguishing each scene individually and identifying targets based on the scenes can improve the accuracy of target recognition.

[0133] A possible implementation method includes: obtaining second data based on the first data, the second data including classification feature parameters of the clustering target, the classification feature parameters including one or more of the speed, circumference, length, width, height, heading angle, distance, and opening angle of the clustering target; processing the first data according to a scene classifier and the second data to obtain a scene classification result of the clustering target.

[0134] The second data is obtained after processing the first data and is used to classify the first data into scenes. The second data can be a single type of data including the target's speed, circumference, length, width, height, heading angle, distance, and opening angle, or it can be composite data including multiple types of the above data, or it can be a composite parameter obtained after calculating the above multiple data and can be used as a basis for classification.

[0135] A possible implementation further includes a first target classifier, the first data includes a first cluster target, and the scene classification result of the first cluster target includes that the first cluster target belongs to a first scene; the first target classifier corresponds to the first scene, and the first target classifier is used to classify the first data.

[0136] Optionally, the first target classifier is a traditional machine learning classifier, a deep learning classifier or a multi-frame fusion classifier, and the traditional machine learning classifier includes: naive Bayes classifier, logistic regression classifier, support vector machine classifier (SVM classifier), random forest classifier (RF classifier), decision tree classifier, adaptive boosting classifier (AdaBoost classifier), extreme gradient boosting classifier (XGBoost classifier); the deep learning classifier includes: fully connected neural network classifier, convolutional neural network classifier, deep residual neural network classifier; the multi-frame fusion classifier includes a multi-frame fusion classifier based on the above-mentioned traditional machine learning classifier or deep learning classifier using Kalman filtering, hidden Markov, and naive Bayes technology for multi-frame joint processing. The classifiers corresponding to different scenarios can be classifiers of the same type or different types, and this application does not impose any restrictions on this.

[0137] The present application provides a terminal, which includes the target recognition device included in any of the above embodiments.

[0138] Optionally, the terminal is a vehicle, a drone or a robot.

[0139] The present application provides a computer-readable storage medium, characterized in that the computer-readable storage medium is used to store instructions, and when the instructions are executed, the method in any one of the above possible implementation methods is implemented.

[0140] Regarding the implementation effects of the detection device, the terminal, and the computer-readable storage medium, reference may be made to the introduction of various implementation methods, which will not be repeated here.

[0141] In the above embodiments, 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 instructions are loaded and executed on a computer, the process or function according to the embodiment of the present application is generated in whole or in part. 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 via a wired (e.g., coaxial cable, optical fiber, digital subscriber line (DSL)) or wireless (e.g., infrared, wireless, microwave, etc.) method. 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 available media integrated. The available medium can be a magnetic medium (e.g., a floppy disk, a hard disk, a tape), an optical medium (e.g., a high-density digital video disc (DVD)), or a semiconductor medium (e.g., a solid state drive (SSD)).

[0142] Those skilled in the art will understand that the various numerical numbers such as first and second involved in this application are only for the convenience of description and are not used to limit the scope or sequence of the embodiments of this application.

[0143] Those skilled in the art will appreciate that the units and algorithm steps of each example described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware, or a combination of computer software and electronic hardware. Whether these functions are performed in hardware or software depends on the specific application and design constraints of the technical solution. Professional and technical personnel can use different methods to implement the described functions for each specific application, but such implementation should not be considered beyond the scope of this application.

[0144] Those skilled in the art will clearly understand that, for the convenience and brevity of description, the specific working processes of the systems, devices and units described above can refer to the corresponding processes in the aforementioned method embodiments and will not be repeated here.

Claims

1. A target recognition method, characterized in that: include: obtaining first data; The first data includes cluster target information obtained based on millimeter wave radar measurement data; The information of the clustered targets includes one or more of sparse point cloud data, distance data, azimuth data, Doppler data, and occupancy grid map data (OGM data) of the clustered targets; The first data is processed according to at least two scene classifiers to obtain a scene classification result of the clustering target.

2. The target recognition method according to claim 1, characterized in that: The scene classification results include: one or more of radial target scene, tangential target scene, traversing target scene, straight target scene, stationary target scene, field of view boundary target scene, occluded target scene, short-range target scene, medium-range target scene, and long-range target scene.

3. The target recognition method according to claim 2, characterized in that: The radial target scene satisfies one or more of the following conditions: the radial velocity of the target is greater than or equal to a first threshold; the tangential velocity of the target is less than or equal to a second threshold; the ratio of the radial velocity of the target to the tangential velocity of the target is greater than or equal to a third threshold; The tangential target scenario satisfies one or more of the following conditions: the target tangential velocity is greater than or equal to a fourth threshold; the target radial velocity is less than or equal to a fifth threshold; the ratio of the target tangential velocity to the radial velocity is greater than or equal to a sixth threshold; The target crossing scenario is a scenario in which the absolute value of the target's heading angle is greater than or equal to the seventh threshold and less than or equal to the eighth threshold; The straight-ahead target scenario is a scenario in which the absolute value of the target's heading angle is greater than or equal to a ninth threshold, or the absolute value of the target's heading angle is less than or equal to a tenth threshold; The stationary target scene satisfies one or more of the following conditions: a target radial velocity is less than or equal to an eleventh threshold and a tangential velocity is less than or equal to a twelfth threshold, and a target velocity is less than or equal to a thirteenth threshold. The field of view boundary target scene is a scene where the target is partially or completely located outside the field of view of the millimeter wave radar; The occluded target scene is a scene in which there is a second target with an opening angle α2 to itself, and there is a first target with an opening angle α1 to itself, α2 and α1 have an overlapping angle β, and β is greater than or equal to the fourteenth threshold; The short-range target scenario is a scenario where the distance between the target and the self is less than or equal to the fifteenth threshold; The medium-distance target scene is a scene where the distance between the target and the self is greater than or equal to the sixteenth threshold and less than or equal to the seventeenth threshold; The long-distance target scenario is a scenario where the distance between the target and the self is greater than or equal to the eighteenth threshold, or a scenario where the distance between the target and the self threshold is greater than or equal to the eighteenth threshold and less than or equal to the nineteenth threshold.

4. The target recognition method according to any one of claims 1 to 3, characterized in that: Also includes, Acquire second data according to the first data, The second data includes classification feature parameters of the clustered targets, and the classification feature parameters include one or more of speed, circumference, length, width, height, heading angle, distance, and opening angle of the clustered targets; The first data is processed according to the scene classifier and the second data to obtain a scene classification result of the clustering target.

5. The target recognition method according to any one of claims 1 to 4, characterized in that: The first data includes a first clustering target, The scene classification result of the first cluster target includes that the first cluster target belongs to a first scene; A first target classifier is used to perform target classification on the first data, where the first target classifier corresponds to the first scene.

6. The target classification and recognition method according to claims 1-5, characterized in that: The first target classifier is a traditional machine learning classifier, a deep learning classifier or a multi-frame fusion classifier, The traditional machine learning classifiers include: Naive Bayes classifier, Logistic Regression classifier, Support Vector Machine classifier (SVM classifier), Random Forest classifier (RF classifier), Decision Tree classifier, Adaptive Boosting classifier (AdaBoost classifier), Extreme Gradient Boosting classifier (XGBoost classifier); The deep learning classifier includes: a fully connected neural network classifier, a convolutional neural network classifier, and a deep residual neural network classifier; The multi-frame fusion classifier includes a multi-frame fusion classifier that uses Kalman filtering, hidden Markov, and naive Bayes technology to perform multi-frame joint processing based on the above-mentioned traditional machine learning classifier or deep learning classifier.

7. The target recognition method according to any one of claims 1 to 6, characterized in that: Also includes: The scene classification result of the first cluster target includes that the first cluster target belongs to at least two scenes, and the first cluster target is classified based on the target classifiers corresponding to the at least two scenes; If the same target has target classification results output in multiple target classifiers, the target classification result of the target is determined after performing a first operation on the multiple target classification results of the same target.

8. The target classification and recognition method according to any one of claims 1 to 7, characterized in that: Also includes: The scene classification result of the first cluster target includes that the first cluster target belongs to at least two scenes, and the first cluster target is classified based on the target classifiers corresponding to the at least two scenes; If the same target has a target classification result output only in a single target classifier, the classification result of the classifier for the target is the final target classification result.

9. A target recognition device, characterized in that: Including acquisition module and processing module, The acquisition module is used to acquire first data, where the first data includes cluster target information obtained based on millimeter wave radar measurement data; the cluster target information includes one or more of sparse point cloud data, distance data, azimuth data, Doppler data, and occupancy grid map data (OGM data) of the cluster target; The processing module is used to process the first data to obtain a scene classification result of the clustering target.

10. The target recognition device according to claim 9, characterized in that: The scene classification results include: One or more of radial target scene, tangential target scene, traversing target scene, straight target scene, stationary target scene, field of view boundary target scene, occluded target scene, short-range target scene, medium-range target scene, and long-range target scene.

11. The target recognition device according to any one of claims 9 to 10, characterized in that: The radial target scene satisfies one or more of the following conditions: the radial velocity of the target is greater than or equal to a first threshold; the tangential velocity of the target is less than or equal to a second threshold; the ratio of the radial velocity of the target to the tangential velocity of the target is greater than or equal to a third threshold; The tangential target scenario satisfies one or more of the following conditions: the target tangential velocity is greater than or equal to a fourth threshold; the target radial velocity is less than or equal to a fifth threshold; the ratio of the target tangential velocity to the radial velocity is greater than or equal to a sixth threshold; The target-crossing scenario is a scenario in which the absolute value of the target's heading angle is greater than or equal to a seventh threshold and less than or equal to an eighth threshold. The straight-ahead target scenario is a scenario in which the absolute value of the target's heading angle is greater than or equal to a ninth threshold, or the absolute value of the target's heading angle is less than or equal to a tenth threshold; The stationary target scene satisfies one or more of the following conditions: a target radial velocity is less than or equal to an eleventh threshold and a tangential velocity is less than or equal to a twelfth threshold, and a target velocity is less than or equal to a thirteenth threshold. The field of view boundary target scene is a scene where the target is partially or completely located outside the field of view of the millimeter wave radar; The occluded target scene is a scene in which there is a second target with an opening angle α2 to itself, and there is a first target with an opening angle α1 to itself, α2 and α1 have an overlapping angle β, and β is greater than or equal to the fourteenth threshold; The short-range target scenario is a scenario where the distance between the target and the self is less than or equal to the fifteenth threshold; The medium-distance target scene is a scene where the distance between the target and the self is greater than or equal to the sixteenth threshold and less than or equal to the seventeenth threshold; The long-distance target scenario is a scenario where the distance between the target and the self is greater than or equal to the eighteenth threshold, or a scenario where the distance between the target and the self threshold is greater than or equal to the eighteenth threshold and less than or equal to the nineteenth threshold.

12. The target recognition device according to any one of claims 9 to 11, characterized in that: The first data is also used to obtain second data, The second data includes classification feature parameters of the clustered targets, and the classification feature parameters include one or more of speed, circumference, length, width, height, heading angle, distance, and opening angle of the clustered targets; The processing module is specifically configured to process the first data according to a scene classifier and second data to obtain a scene classification result of a clustering target.

13. The target recognition device according to any one of claims 9 to 12, characterized in that: Also included is a first target classifier, The first data includes a first clustering target, The scene classification result of the first cluster target includes that the first cluster target belongs to a first scene; The first target classifier is used to perform target classification on the first data, and the first classifier corresponds to the first scene.

14. The target recognition device according to claims 9-13, characterized in that: The first target classifier is a traditional machine learning classifier, a deep learning classifier or a multi-frame fusion classifier, The traditional machine learning classifiers include: Naive Bayes classifier, Logistic Regression classifier, Support Vector Machine classifier (SVM classifier), Random Forest classifier (RF classifier), Decision Tree classifier, Adaptive Boosting classifier (AdaBoost classifier), Extreme Gradient Boosting classifier (XGBoost classifier); The deep learning classifier includes: a fully connected neural network classifier, a convolutional neural network classifier, and a deep residual neural network classifier; The multi-frame fusion classifier includes a multi-frame fusion classifier that uses Kalman filtering, hidden Markov, and naive Bayes technology to perform multi-frame joint processing based on the above-mentioned traditional machine learning classifier or deep learning classifier.

15. The target recognition device according to any one of claims 9 to 14, characterized in that: The scene classification result of the first cluster target includes that the first cluster target belongs to at least two scenes, and the first cluster target is classified based on the target classifiers corresponding to the at least two scenes; If the same target has target classification results output in multiple target classifiers, the target classification result of the target is determined after performing a first operation on the multiple target classification results of the same target.

16. The target recognition device according to any one of claims 9 to 15, characterized in that: The scene classification result of the first cluster target includes that the first cluster target belongs to at least two scenes, and the first cluster target is classified based on the target classifiers corresponding to the at least two scenes; If the same target has a target classification result output only in a single target classifier, the classification result of the classifier for the target is the final target classification result.

17. A terminal, characterized in that: The terminal includes the target identification device according to any one of claims 9 to 16. The terminal according to claim 17 , wherein: The terminal is a vehicle, a drone or a robot.

19. A computer-readable storage medium, characterized in that The computer-readable storage medium is used to store instructions, and when the instructions are executed, the method according to any one of claims 1 to 8 is implemented.

Images