Multi-path identification method and related equipment thereof

Abstract

The invention provides a multipath identification method and related equipment thereof. The method comprises the following steps: acquiring radar point cloud data under various scenes frame by frame as original point cloud data to be identified; clustering the original point cloud data to obtain a class cluster result, and associating the class cluster result of the current frame with the class cluster result of the previous frame to obtain a class cluster queue; combining the multi-dimensional feature units corresponding to all the class cluster results of the current frame to obtain a multi-dimensional feature information group; inputting the information of the multi-dimensional feature information group of the current frame into a pre-trained AdaBoost model, and outputting a multi-path judgment result of each class cluster result; and if the number of times that the class cluster result in the class cluster queue is marked as multipath in the continuous N frames exceeds a first preset threshold value, determining that a target corresponding to the class cluster result is a multipath target. The multipath identification method of the scheme can cope with the identification of multiple scenes and complex scenes, reduces false alarms, and improves the accuracy of multipath identification.

Description

Technical Field

[0001] The present invention relates to the field of radar, and in particular to a multipath identification method and related equipment. Background Art

[0002] Due to its advantages such as long range and strong penetration, vehicle-mounted millimeter-wave radar is widely used in the field of modern automobile autonomous driving. When working, the radar transmits and receives electromagnetic waves, and senses the environmental conditions of the target area by extracting information from the echo signal. However, due to complex road conditions and physical limitations, millimeter-wave radar will cause many problems when used, and multipath is one of them. Figure 1 The common propagation paths of multipath phenomena are shown in the figure. The propagation paths of such multipaths are clear, and the characteristics of multipath false points are obvious. They can be easily judged based on characteristic values ​​such as distance, speed, and angle. However, this conventional multipath judgment method can only detect multipath false targets in such simple scenes. It requires clear strong reflectors in the scene, and it is necessary to construct geometric models of multiple possible scenes, and then compare them one by one. For relatively complex scenes or scenes with unclear strong reflectors, traditional multipath judgment methods are more difficult to identify and judge. The false targets generated by the multipath problem will interfere with the radar's target detection, causing false alarms and other problems, which seriously affects the automatic driving of the car.

[0003] At present, the more common multipath identification method is to judge false targets based on geometric models. This type of method relies on the geometric representation of the electromagnetic wave propagation path, summarizes the electromagnetic wave propagation paths that may produce multipath, and marks the suspected multipath targets accordingly. However, in actual applications, due to the complex and changeable road conditions, this multipath identification method based on geometric modeling is not ideal, and this type of method has relatively strict conditions and often requires clear reflection boundaries in the actual environment, so the scope of application is relatively narrow.

[0004] Therefore, in actual scenarios, due to the complex road environment and various multipath propagation paths, existing methods are difficult to identify and suppress multipath interference. Summary of the invention

[0005] In view of this, an embodiment of the present invention provides a passenger identification method and related equipment, aiming to solve the technical problem of low multipath identification accuracy of existing vehicle-mounted radar.

[0006] In a first aspect, an embodiment of the present invention provides a multipath identification method, including:

[0007] Acquire radar point cloud data in a variety of scenes frame by frame, and use the radar point cloud data as original point cloud data to be identified, wherein the various scenes include at least one of a tunnel scene, a congested scene, an open scene, a straight scene, a curved scene, and an iron guardrail scene;

[0008] Based on the density clustering algorithm, the original point cloud data is clustered to obtain cluster results, and the cluster results of the current frame are associated with the cluster results of the previous frame. The cluster results of all frames are associated to obtain a cluster queue;

[0009] The various cluster results of the current frame are subjected to multi-dimensional feature extraction according to a preset algorithm and are taken as a unit, and the multi-dimensional feature units corresponding to all the cluster results of the current frame are combined to obtain a multi-dimensional feature information group;

[0010] Input the information of the multi-dimensional feature information group of the current frame into the pre-trained AdaBoost model, and output the multipath judgment results of each cluster result;

[0011] If the number of times that the cluster results in the cluster queue are marked as multipath in N consecutive frames exceeds a first preset threshold, it is determined that the target corresponding to the cluster result is a multipath target.

[0012] Preferably, the steps of clustering the original point cloud data based on a density clustering algorithm to obtain cluster results, and associating the cluster results of the current frame with the cluster results of the previous frame include:

[0013] Based on the density clustering algorithm, the original point cloud data is clustered to obtain a cluster result, and the parameters of the density clustering algorithm include at least one of distance, speed and angle;

[0014] The parameter values ​​of each point cloud in the cluster result are weighted averaged to obtain the parameter values ​​of the cluster result. The point cloud with a larger Mag value has a greater weight.

[0015] The cluster results are sorted by distance, and the parameters are compensated in combination with the vehicle body information. The binary search algorithm is used to search all the cluster results of the previous frame one by one.

[0016] Mark all second cluster results whose distance difference with the first cluster result is less than a second preset threshold, to obtain a third cluster result, wherein the first cluster result is included in all cluster results of the current frame, and the second cluster result is included in all cluster results of the previous frame;

[0017] The Euclidean distance between the traversed third-category cluster results and the first-category cluster results is calculated one by one, and the third-category cluster result with the smallest Euclidean distance is recorded, and the first-category cluster result is associated with the third-category cluster result with the smallest Euclidean distance.

[0018] Preferably, the steps of calculating the Euclidean distances between the traversed third-category cluster results and the first-category cluster results one by one, recording the third-category cluster result with the smallest Euclidean distance, and associating the first-category cluster result with the third-category cluster result with the smallest Euclidean distance include:

[0019] At the same time, speed and angle pruning operations are combined during the traversal process to save time.

[0020] Preferably, the calculation formula of the Euclidean distance is:

[0021]

[0022] Among them, d represents the Euclidean distance, △R represents the distance difference of cluster results, △V represents the speed difference of cluster results, and △A represents the angle difference of cluster results.

[0023] Preferably, the multidimensional features include the number, distance, speed, angle, Mag value, signal-to-noise ratio, radar cross section, transverse coordinate, and longitudinal coordinate of the point cloud in the cluster results, the weighted average distance, weighted average speed, weighted average angle, weighted average Mag value, weighted average signal-to-noise ratio, weighted average radar cross section, weighted average transverse coordinate, and weighted average longitudinal coordinate of the point cloud in the cluster results, and any of the distance variance, speed variance, angle variance, Mag value variance, signal-to-noise ratio variance, radar cross section variance, transverse coordinate variance, and longitudinal coordinate variance of the point cloud in the cluster results.

[0024] Preferably, after the step of inputting the information of the multidimensional feature information group of the current frame into the pre-trained AdaBoost model and outputting the multipath judgment results of each cluster result, the method further comprises:

[0025] According to the associated cluster results, a cluster queue is formed. After the cluster result of each frame is determined, the cluster results in the cluster queue are pushed and popped, and the number of frames n determined as multipath targets in the cluster queue is updated at the same time.

[0026] If n is greater than the first threshold, the target corresponding to the cluster result is determined to be a multipath target.

[0027] Preferably, the training method of the AdaBoost model includes:

[0028] Acquire radar point cloud data in a variety of scenes frame by frame, and use the radar point cloud data as training point cloud data, wherein the various scenes include at least one of a tunnel scene, a congested scene, an open scene, a straight scene, a curved scene, and an iron guardrail scene;

[0029] Based on the density clustering algorithm, the training point cloud data is clustered to obtain cluster results, and the cluster results of the current frame are associated with the cluster results of the previous frame;

[0030] The various cluster results of the current frame are subjected to multi-dimensional feature extraction according to a preset algorithm and are taken as a unit. The multi-dimensional feature units corresponding to all cluster results of the current frame are combined to obtain a multi-dimensional feature information group, and a label is added to each cluster result to mark whether the target corresponding to the cluster result is a multipath target;

[0031] Construct a feature data set based on a large number of labeled multi-dimensional feature information groups, and train the AdaBoost model based on the feature data set;

[0032] During the training process, the AdaBoost algorithm is used to iteratively train multiple SVM base classifiers. In each iteration, the AdaBoost algorithm adjusts the sample weights according to the results of the previous iteration, so that the AdaBoost model pays more attention to samples with high error rates.

[0033] In a second aspect, the present invention provides a multipath identification device, comprising:

[0034] A signal processing module, used for acquiring radar point cloud data in various scenes frame by frame, and using the radar point cloud data as original point cloud data to be identified, wherein the various scenes include at least one of a tunnel scene, a congested scene, an open scene, a straight scene, a curved scene, and an iron guardrail scene;

[0035] The clustering module is used to cluster the original point cloud data based on the density clustering algorithm to obtain cluster results, and associate the cluster results of the current frame with the cluster results of the previous frame, perform association operations on the cluster results of all frames, and obtain a cluster queue;

[0036] A feature extraction module is used to extract multi-dimensional features from each cluster result of the current frame according to a preset algorithm and combine the multi-dimensional feature units corresponding to all cluster results of the current frame as a unit to obtain a multi-dimensional feature information group;

[0037] A data preprocessing module is used to input the information of the multi-dimensional feature information group of the current frame into the pre-trained AdaBoost model and output the multipath judgment results of each cluster result;

[0038] The multipath detection module is used to determine that the target corresponding to the cluster result is a multipath target if the number of times that the cluster result in the cluster queue is marked as multipath in N consecutive frames exceeds a first preset threshold.

[0039] In a third aspect, the present invention provides an electronic device, comprising: at least one processor, at least one memory, and computer program instructions stored in the memory, wherein when the computer program instructions are executed by the processor, the method as described above is implemented.

[0040] In a fourth aspect, the present invention provides a storage medium having computer program instructions stored thereon, wherein the method described above is implemented when the computer program instructions are executed by a processor.

[0041] In summary, the beneficial effects of the present invention are as follows:

[0042] The multipath identification method and related equipment provided by the embodiment of the present invention classify radar point cloud data according to multi-dimensional features and use a machine learning model to judge the data. The multipath identification method of this solution can cope with the identification of multiple scenes and complex scenes, reduce false alarms, and improve the accuracy of multipath identification. In addition, this method does not need to establish a geometric model corresponding to multiple scenes, which can reduce the cost of the multipath identification solution. BRIEF DESCRIPTION OF THE DRAWINGS

[0043] In order to more clearly illustrate the technical solution of the embodiment of the present invention, the drawings required for use in the embodiment of the present invention will be briefly introduced below. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying any creative work, and these are all within the protection scope of the present invention.

[0044] Figure 1 Schematic diagram of multipath phenomenon.

[0045] Figure 2 The figure is a flow chart of a first embodiment of a multipath identification method of the present invention.

[0046] Figure 3 A schematic diagram of a cluster queue.

[0047] Figure 4 The figure is a schematic diagram of the structure of a multipath identification device of the present invention.

[0048] Figure 5 It is a schematic diagram of the structure of an electronic device according to an embodiment of the present invention. DETAILED DESCRIPTION

[0049] The features and exemplary embodiments of various aspects of the present invention will be described in detail below. In order to make the purpose, technical solutions and advantages of the present invention clearer, the present invention will be further described in detail below in conjunction with the accompanying drawings and Examples. It should be understood that the specific embodiments described herein are only configured to explain the present invention and are not configured to limit the present invention. For those skilled in the art, the present invention can be implemented without the need for some of these specific details. The following description of the embodiments is only to provide a better understanding of the present invention by illustrating examples of the present invention.

[0050] It should be noted that, in this article, relational terms such as first and second, etc. are only used to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Moreover, the terms "include", "comprise" or any other variants thereof are intended to cover non-exclusive inclusion, so that a process, method, article or device including a series of elements includes not only those elements, but also other elements not explicitly listed, or also includes elements inherent to such process, method, article or device. In the absence of further restrictions, the elements defined by the statement "include..." do not exclude the presence of other identical elements in the process, method, article or device including the elements.

[0051] Reference Figure 2 and 3 The present invention provides a multipath identification method, comprising:

[0052] S1: acquiring radar point cloud data in a variety of scenes frame by frame, and using the radar point cloud data as original point cloud data to be identified, wherein the various scenes include at least one of a tunnel scene, a congested scene, an open scene, a straight scene, a curved scene, and an iron guardrail scene;

[0053] S2: Based on the density clustering algorithm, the original point cloud data is clustered to obtain cluster results, and the cluster results of the current frame are associated with the cluster results of the previous frame, and the cluster results of all frames are associated to form a cluster queue;

[0054] S3: extracting multidimensional features from each cluster result of the current frame according to a preset algorithm and treating them as a unit, combining the multidimensional feature units corresponding to all cluster results of the current frame to obtain a multidimensional feature information group;

[0055] S4: inputting the information of the multi-dimensional feature information group of the current frame into the pre-trained AdaBoost model, and outputting the multipath judgment results of each cluster result;

[0056] S5: If the number of times that the cluster results in the cluster queue are marked as multipath in N consecutive frames exceeds a first preset threshold, it is determined that the target corresponding to the cluster result is a multipath target.

[0057] In an embodiment of the present invention, a multipath identification system acquires radar point cloud data in a variety of scenarios frame by frame, and uses these data as original point cloud data to be identified, wherein the various scenarios include at least one of tunnel scenes, congested scenes, open scenes, straight scenes, curve scenes, and iron guardrail scenes. For the radar point cloud data of the current frame, the original point cloud data of the current frame is first clustered based on a density clustering algorithm, and in the clustering process, the overall eigenvalues ​​of the cluster results are calculated, and the eigenvalues ​​include but are not limited to parameters such as distance, speed, angle, and signal-to-noise ratio. Exemplarily, there are targets A, B, and C in the current frame, and based on the clustering algorithm, the corresponding cluster results A, B, and C are obtained. The cluster results A, B, and C of the current frame are associated with the cluster results A, B, and C of the previous frame, and the cluster results of all frames are associated to obtain cluster queues A, B, and C, respectively. The cluster queues are as follows: Figure 3 As shown. The cluster results of the current frame are subjected to multi-dimensional feature extraction according to a preset algorithm and are used as a unit. The multi-dimensional feature units corresponding to all cluster results of the current frame are combined to obtain a multi-dimensional feature information group. Exemplarily, the cluster results A, B and C are subjected to multi-dimensional feature extraction according to a preset algorithm to obtain corresponding feature units A, B and C. Feature units A, B and C are combined to obtain a multi-dimensional feature information group. The information of the multi-dimensional feature information group of the current frame is input into the pre-trained AdaBoost model, and the multi-path judgment results of each cluster result are output, as shown. Figure 3 As shown, according to the multipath detection results of the cluster results in the cluster queue, if the number of times the cluster results are marked as multipath in N consecutive frames exceeds the first preset threshold, the target corresponding to the cluster result is determined to be a multipath target. Exemplarily, the multidimensional feature information group of the current frame is input into the Adaboost model, and the model determines the multipath judgment results of the cluster results A, B and C of the current frame. The cluster results of the cluster queues A, B and C perform multipath detection judgment in turn. If the number of times the cluster results are marked as multipath in N consecutive frames exceeds the first preset threshold, the target corresponding to the cluster result is determined to be a multipath target. According to the classification of radar point cloud data based on multidimensional features, the machine learning model is used to judge the data. The multipath identification method of this scheme can cope with the identification of multiple scenes and complex scenes, reduce false alarms, and improve the accuracy of multipath identification. In addition, this method does not need to establish geometric models corresponding to multiple scenes, which can reduce the cost of the multipath identification scheme.

[0058] Furthermore, based on a density clustering algorithm, the original point cloud data is clustered to obtain a cluster result, and the cluster result of the current frame is associated with the cluster result of the previous frame, step S2, including:

[0059] S21: clustering the original point cloud data based on a density clustering algorithm to obtain a cluster result, wherein the parameters of the density clustering algorithm include at least one of distance, speed and angle;

[0060] S22: weighted average the parameter values ​​of each point cloud in the cluster result to obtain the parameter values ​​of the cluster result, wherein the point cloud with a larger Mag value has a larger weight;

[0061] S23: sorting the cluster results according to the distance, performing parameter compensation in combination with the vehicle body information, and using a binary search algorithm to search for all cluster results of the previous frame one by one;

[0062] S24: Mark all second cluster results whose distance difference with the first cluster result is less than a second preset threshold, to obtain a third cluster result, wherein the first cluster result is included in all cluster results of the current frame, and the second cluster result is included in all cluster results of the previous frame;

[0063] S25: Calculate the Euclidean distances between the traversed third cluster results and the first cluster results one by one, record the third cluster result with the smallest Euclidean distance, and associate the first cluster result with the third cluster result with the smallest Euclidean distance.

[0064] In an embodiment of the present invention, the original point cloud data is clustered based on a density clustering algorithm to obtain a cluster result, and the parameters of the density clustering algorithm include at least one of distance, speed and angle; the parameter values ​​of each point cloud in the cluster result are weighted averaged to obtain the parameter values ​​of the cluster result, wherein the parameter values ​​include but are not limited to distance, speed, angle, signal-to-noise ratio, and the point cloud with a larger Mag value has a larger weight. The cluster results are sorted according to distance, and parameter compensation is performed in combination with vehicle body information. A binary search algorithm is used to search all cluster results of the previous frame one by one. Specifically, in the association process, the cluster distance, cluster speed, and cluster angle are compensated in combination with vehicle body information, frame period and other data, and matching is performed based on the binary search algorithm with the minimum Euclidean distance as the criterion. The characteristic value distance of the cluster results of the two frames before and after compensation is:

[0065]

[0066] Among them, △R represents the distance difference of cluster results, R cur Represents the distance of the cluster result of the current frame, R pre Indicates the distance of the cluster result of the previous frame, V cur Indicates the speed of the cluster result of the current frame, V pre Indicates the speed of the clustering result of the previous frame;

[0067] ΔV=V cur -V pre ;

[0068] Among them, △V represents the speed difference of cluster results, V cur Indicates the speed of the cluster result of the current frame, V pre Indicates the speed of the clustering result of the previous frame;

[0069] ΔA=A cur -A pre -YawDeg

[0070] Among them, A cur Indicates the angle of the cluster result of the current frame, A pre Represents the angle of the cluster result of the previous frame, and YawDeg represents the steering angle of the vehicle. Mark all second cluster results whose distance difference with the first cluster result is less than the second preset threshold, and obtain the third cluster result, wherein the first cluster result is included in all cluster results of the current frame, and the second cluster result is included in all cluster results of the previous frame. Exemplarily, cluster result A1 of the current frame is the first cluster result, cluster result A0, cluster result B0, and cluster result C0 are the cluster results of the previous frame, that is, the second cluster result. Cluster result A0 and cluster result B0 are both cluster results whose distance difference of cluster results is less than the second preset threshold. Mark cluster result A0 and cluster result B0 to obtain the third cluster result. Calculate the Euclidean distance of the traversed third cluster results and the first cluster results one by one, and record the third cluster result with the smallest Euclidean distance, and associate the first cluster result with the third cluster result with the smallest Euclidean distance. Exemplarily, the Euclidean distances between cluster result A1 and cluster results A0 and B0 are calculated by traversing, and if the Euclidean distance between cluster result A0 and cluster result A1 is the smallest, cluster results A1 and A0 are associated.

[0071] Furthermore, speed and angle pruning operations are combined during the traversal process to save time overhead.

[0072] Furthermore, the calculation formula of Euclidean distance is:

[0073]

[0074] Among them, d represents the Euclidean distance, △R represents the distance difference of cluster results, △V represents the speed difference of cluster results, and △A represents the angle difference of cluster results.

[0075] Furthermore, the multidimensional features include the number, distance, speed, angle, Mag value, signal-to-noise ratio, radar cross section, transverse coordinate, and longitudinal coordinate of the point cloud in the cluster results, the weighted average distance, weighted average speed, weighted average angle, weighted average Mag value, weighted average signal-to-noise ratio, weighted average radar cross section, weighted average transverse coordinate, and weighted average longitudinal coordinate of the point cloud in the cluster results, and any of the distance variance, speed variance, angle variance, Mag value variance, signal-to-noise ratio variance, radar cross section variance, transverse coordinate variance, and longitudinal coordinate variance of the point cloud in the cluster results.

[0076] like Figure 3 As shown, after inputting the information of the multi-dimensional feature information group of the current frame into the pre-trained AdaBoost model and outputting the multipath judgment results of each cluster result in step S4, the method includes:

[0077] S50: forming a cluster queue according to the associated cluster results, and after judging the cluster results of each frame, performing stacking and popping operations on the cluster results in the cluster queue, and updating the number of frames n judged as multipath targets in the cluster queue at the same time;

[0078] S51: If n is greater than a first threshold, it is determined that the target corresponding to the cluster result is a multipath target.

[0079] In the embodiment of the present invention, a cluster queue is formed according to the associated cluster result, and after the cluster result of each frame is determined, the cluster result in the cluster queue is pushed and popped, and the number of frames n determined as multipath targets in the cluster queue is updated. If the number of frames of multipath targets accumulated in the cluster queue is greater than the first threshold, the target corresponding to the cluster queue is also determined to be a multipath target.

[0080] Furthermore, the training method of the AdaBoost model includes:

[0081] Sa: acquiring radar point cloud data in various scenes frame by frame, and using the radar point cloud data as training point cloud data, wherein the various scenes include at least one of a tunnel scene, a congested scene, an open scene, a straight scene, a curved scene, and an iron guardrail scene;

[0082] Sb: Based on the density clustering algorithm, the training point cloud data is clustered to obtain cluster results, and the cluster results of the current frame are associated with the cluster results of the previous frame;

[0083] Sc: extract multidimensional features from each cluster result of the current frame according to a preset algorithm and treat them as a unit, combine the multidimensional feature units corresponding to all the cluster results of the current frame to obtain a multidimensional feature information group, and add a label to each cluster result to mark whether the target corresponding to the cluster result is a multipath target;

[0084] Sd: constructs a feature data set based on a large number of labeled multidimensional feature information groups, and trains the AdaBoost model based on the feature data set;

[0085] Se: During the training process, the AdaBoost algorithm is used to iteratively train multiple SVM base classifiers. In each iteration, the AdaBoost algorithm adjusts the sample weights according to the results of the previous iteration, so that the AdaBoost model pays more attention to samples with high error rates.

[0086] In an embodiment of the present invention, the scheme will label the data for model training to distinguish whether the clustering result is a multipath target. In the AdaBoost algorithm, the system iteratively trains multiple base classifiers, each of which is a support vector machine (SVM). SVM is a powerful classification algorithm that finds an optimal hyperplane in the feature space to distinguish different categories. In each iteration, the AdaBoost algorithm adjusts the weight of the sample according to the result of the previous iteration. Specifically, if a sample is misclassified in the previous iteration, its weight will increase; if a sample is correctly classified, its weight will decrease. The purpose of this is to allow the model to pay more attention to those samples that are difficult to classify (i.e., samples with high error rates) in subsequent iterations. By adjusting the sample weights, the AdaBoost model can pay more attention to those samples with high error rates, thereby improving the overall performance of the model. This is because in ensemble learning, improving the performance of the model on difficult samples can usually significantly improve the overall classification effect.

[0087] Reference Figure 4 The present invention provides a multipath identification device, comprising:

[0088] The signal processing module 1 is used to obtain radar point cloud data in various scenes frame by frame, and use the radar point cloud data as the original point cloud data to be identified, wherein the various scenes include at least one of a tunnel scene, a congested scene, an open scene, a straight scene, a curved scene, and an iron guardrail scene;

[0089] Clustering module 2 is used to cluster the original point cloud data based on the density clustering algorithm to obtain cluster results, and associate the cluster results of the current frame with the cluster results of the previous frame, perform an association operation on the cluster results of all frames, and obtain a cluster queue;

[0090] Feature extraction module 3, used to extract multi-dimensional features from each cluster result of the current frame according to a preset algorithm and combine the multi-dimensional feature units corresponding to all cluster results of the current frame as a unit to obtain a multi-dimensional feature information group;

[0091] The data preprocessing module 4 is used to input the information of the multi-dimensional feature information group of the current frame into the pre-trained AdaBoost model, and output the multipath judgment results of each cluster result;

[0092] The multipath detection module 5 is used to determine that the target corresponding to the cluster result is a multipath target if the number of times that the cluster result in the cluster queue is marked as multipath in N consecutive frames exceeds a first preset threshold.

[0093] In an embodiment of the present invention, a multipath identification system acquires radar point cloud data in a variety of scenarios frame by frame, and uses these data as original point cloud data to be identified, wherein the various scenarios include at least one of tunnel scenes, congested scenes, open scenes, straight scenes, curve scenes, and iron guardrail scenes. For the radar point cloud data of the current frame, the original point cloud data of the current frame is first clustered based on a density clustering algorithm, and in the clustering process, the overall eigenvalues ​​of the cluster results are calculated, and the eigenvalues ​​include but are not limited to parameters such as distance, speed, angle, and signal-to-noise ratio. Exemplarily, there are targets A, B, and C in the current frame, and based on the clustering algorithm, the corresponding cluster results A, B, and C are obtained. The cluster results A, B, and C of the current frame are associated with the cluster results A, B, and C of the previous frame, and the cluster results of all frames are associated to obtain cluster queues A, B, and C, respectively. The cluster queues are as follows: Figure 3 As shown. The cluster results of the current frame are subjected to multi-dimensional feature extraction according to a preset algorithm and are used as a unit. The multi-dimensional feature units corresponding to all cluster results of the current frame are combined to obtain a multi-dimensional feature information group. Exemplarily, the cluster results A, B and C are subjected to multi-dimensional feature extraction according to a preset algorithm to obtain corresponding feature units A, B and C. Feature units A, B and C are combined to obtain a multi-dimensional feature information group. The information of the multi-dimensional feature information group of the current frame is input into the pre-trained AdaBoost model, and the multi-path judgment results of each cluster result are output, as shown. Figure 3As shown, according to the multipath detection results of the cluster results in the cluster queue, if the number of times the cluster results are marked as multipath in N consecutive frames exceeds the first preset threshold, the target corresponding to the cluster result is determined to be a multipath target. Exemplarily, the multidimensional feature information group of the current frame is input into the Adaboost model, and the model determines the multipath judgment results of the cluster results A, B and C of the current frame. The cluster results of the cluster queues A, B and C perform multipath detection judgment in turn. If the number of times the cluster results are marked as multipath in N consecutive frames exceeds the first preset threshold, the target corresponding to the cluster result is determined to be a multipath target. According to the classification of radar point cloud data based on multidimensional features, the machine learning model is used to judge the data. The multipath identification method of this scheme can cope with the identification of multiple scenes and complex scenes, reduce false alarms, and improve the accuracy of multipath identification. In addition, this method does not need to establish geometric models corresponding to multiple scenes, which can reduce the cost of the multipath identification scheme.

[0094] In addition, combined Figure 5 The multipath identification method described in the embodiment of the present invention may be implemented by an electronic device. Figure 5 A schematic diagram of the hardware structure of an electronic device provided by an embodiment of the present invention is shown.

[0095] The electronic device may include a processor and a memory storing computer program instructions.

[0096] Specifically, the processor may include a central processing unit (CPU), or an application specific integrated circuit (ASIC), or may be configured to implement one or more integrated circuits of the embodiments of the present invention.

[0097] The memory may include a large capacity memory for data or instructions. By way of example and not limitation, the memory may include a hard disk drive (HDD), a floppy disk drive, a flash memory, an optical disk, a magneto-optical disk, a magnetic tape, or a universal serial bus (USB) drive or a combination of two or more of these. Where appropriate, the memory may include a removable or non-removable (or fixed) medium. Where appropriate, the memory may be inside or outside a data processing device. In a particular embodiment, the memory is a non-volatile solid-state memory. In a particular embodiment, the memory includes a read-only memory (ROM). Where appropriate, the ROM may be a mask-programmed ROM, a programmable ROM (PROM), an erasable PROM (EPROM), an electrically erasable PROM (EEPROM), an electrically rewritable ROM (EAROM) or a flash memory or a combination of two or more of these.

[0098] The processor implements any one of the multipath identification methods in the above embodiments by reading and executing computer program instructions stored in the memory.

[0099] In one example, the electronic device may further include a communication interface and a bus. Figure 5 As shown, the processor, memory, and communication interface are connected via a bus and communicate with each other.

[0100] The communication interface is mainly used to implement communication between the modules, devices, units and / or equipment in the embodiments of the present invention.

[0101] Bus includes hardware, software or both, and the parts of electronic equipment are coupled to each other.For example, but not limitation, bus may include accelerated graphics port (AGP) or other graphics bus, enhanced industrial standard architecture (EISA) bus, front side bus (FSB), hypertransport (HT) interconnection, industrial standard architecture (ISA) bus, infinite bandwidth interconnection, low pin count (LPC) bus, memory bus, micro channel architecture (MCA) bus, peripheral component interconnection (PCI) bus, PCI-Express (PCI-X) bus, serial advanced technology attachment (SATA) bus, video electronics standard association local (VLB) bus or other suitable bus or two or more of these combinations. In suitable cases, bus may include one or more buses. Although the embodiment of the present invention describes and shows a specific bus, the present invention considers any suitable bus or interconnection.

[0102] In addition, the embodiments of the present invention may be implemented by providing a computer-readable storage medium having computer program instructions stored thereon; when the computer program instructions are executed by a processor, any multipath identification method in the above embodiments is implemented.

[0103] In summary, the embodiments of the present invention provide a multipath identification method and related devices.

[0104] It should be clear that the present invention is not limited to the specific configuration and processing described above and shown in the figures. For the sake of simplicity, a detailed description of the known method is omitted here. In the above embodiments, several specific steps are described and shown as examples. However, the method process of the present invention is not limited to the specific steps described and shown, and those skilled in the art can make various changes, modifications and additions, or change the order between the steps after understanding the spirit of the present invention.

[0105] The functional blocks shown in the above-described block diagram can be implemented as hardware, software, firmware or a combination thereof. When implemented in hardware, it can be, for example, an electronic circuit, an application specific integrated circuit (ASIC), appropriate firmware, a plug-in, a function card, etc. When implemented in software, the elements of the present invention are programs or code segments that are used to perform the required tasks. The program or code segment can be stored in a machine-readable medium, or transmitted on a transmission medium or a communication link by a data signal carried in a carrier wave. "Machine-readable medium" can include any medium capable of storing or transmitting information. Examples of machine-readable media include electronic circuits, semiconductor memory devices, ROM, flash memory, erasable ROM (EROM), floppy disks, CD-ROMs, optical disks, hard disks, optical fiber media, radio frequency (RF) links, etc. The code segment can be downloaded via a computer network such as the Internet, an intranet, etc.

[0106] It should also be noted that the exemplary embodiments mentioned in the present invention describe some methods or systems based on a series of steps or devices. However, the present invention is not limited to the order of the above steps, that is, the steps can be performed in the order mentioned in the embodiments, or in a different order from the embodiments, or several steps can be performed simultaneously.

[0107] The above is only a specific implementation of the present invention. Those skilled in the art can clearly understand that for the convenience and simplicity of description, the specific working processes of the systems, modules and units described above can refer to the corresponding processes in the aforementioned method embodiments, and will not be repeated here. It should be understood that the protection scope of the present invention is not limited to this. Any technician familiar with the technical field can easily think of various equivalent modifications or replacements within the technical scope disclosed by the present invention, and these modifications or replacements should be covered within the protection scope of the present invention.

Claims

1. A multipath identification method, characterized in that: include: Acquire radar point cloud data in a variety of scenes frame by frame, and use the radar point cloud data as original point cloud data to be identified, wherein the various scenes include at least one of a tunnel scene, a congested scene, an open scene, a straight scene, a curved road scene, and an iron guardrail scene; Based on the density clustering algorithm, the original point cloud data is clustered to obtain cluster results, and the cluster results of the current frame are associated with the cluster results of the previous frame, and the cluster results of all frames are associated to obtain a cluster queue; Extracting multidimensional features from each cluster result of the current frame according to a preset algorithm and treating them as a unit, combining the multidimensional feature units corresponding to all the cluster results of the current frame to obtain a multidimensional feature information group; Inputting the information of the multidimensional feature information group of the current frame into a pre-trained AdaBoost model, and outputting the multipath judgment results of each cluster result; If the number of times that the cluster result in the cluster queue is marked as multipath in N consecutive frames exceeds a first preset threshold, it is determined that the target corresponding to the cluster result is a multipath target.

2. The multipath identification method according to claim 1, characterized in that: The step of clustering the original point cloud data based on the density clustering algorithm to obtain a cluster result, and associating the cluster result of the current frame with the cluster result of the previous frame includes: Based on a density clustering algorithm, clustering the original point cloud data to obtain a cluster result, wherein the parameters of the density clustering algorithm include at least one of distance, speed and angle; Taking a weighted average of the parameter values ​​of each point cloud in the cluster result, the parameter values ​​of each point cloud in the cluster result are obtained, wherein the point cloud with a larger Mag value has a larger weight; The cluster results are sorted according to the distance, parameter compensation is performed in combination with the vehicle body information, and a binary search algorithm is used to search all the cluster results of the previous frame one by one; Mark all second cluster results whose distance difference with the first cluster result is less than a second preset threshold, to obtain a third cluster result, wherein the first cluster result is included in all cluster results of the current frame, and the second cluster result is included in all cluster results of the previous frame; The Euclidean distances between the traversed third cluster results and the first cluster results are calculated one by one, and the third cluster result with the smallest Euclidean distance is recorded, and the first cluster result is associated with the third cluster result with the smallest Euclidean distance.

3. The multipath identification method according to claim 2, characterized in that: The step of calculating the Euclidean distances between the traversed third cluster results and the first cluster results one by one, recording the third cluster result with the smallest Euclidean distance, and associating the first cluster result with the third cluster result with the smallest Euclidean distance comprises: At the same time, speed and angle pruning operations are combined during the traversal process to save time.

4. The multipath identification method according to claim 2 or 3, characterized in that: The calculation formula of the Euclidean distance is: Among them, d represents the Euclidean distance, △R represents the distance difference of cluster results, △V represents the speed difference of cluster results, and △A represents the angle difference of cluster results.

5. The multipath identification method according to any one of claims 1 to 4, characterized in that: The multidimensional features include the number, distance, speed, angle, Mag value, signal-to-noise ratio, radar cross section, transverse coordinate, and longitudinal coordinate of the point cloud in the cluster results, the weighted average distance, weighted average speed, weighted average angle, weighted average Mag value, weighted average signal-to-noise ratio, weighted average radar cross section, weighted average transverse coordinate, and weighted average longitudinal coordinate of the point cloud in the cluster results, and any of the distance variance, speed variance, angle variance, Mag value variance, signal-to-noise ratio variance, radar cross section variance, transverse coordinate variance, and longitudinal coordinate variance of the point cloud in the cluster results.

6. The multipath identification method according to claim 5, characterized in that: After the step of inputting the information of the multidimensional feature information group of the current frame into the pre-trained AdaBoost model and outputting the multipath judgment results of each cluster result, the method further comprises: According to the cluster results after association, the cluster queue is formed, and after the cluster result of each frame is determined, the cluster results in the cluster queue are pushed and popped, and the number of frames n determined as multipath targets in the cluster queue is updated at the same time; If n is greater than the first threshold, it is determined that the target corresponding to the cluster result is a multipath target.

7. The multipath identification method according to claim 6, characterized in that: The training method of the AdaBoost model comprises: Acquire radar point cloud data in a variety of scenes frame by frame, and use the radar point cloud data as training point cloud data, wherein the various scenes include at least one of a tunnel scene, a congested scene, an open scene, a straight scene, a curved road scene, and an iron guardrail scene; Based on a density clustering algorithm, clustering the training point cloud data to obtain a cluster result, and associating the cluster result of the current frame with the cluster result of the previous frame; Extracting multidimensional features from each cluster result of the current frame according to a preset algorithm and treating them as a unit, combining the multidimensional feature units corresponding to all the cluster results of the current frame to obtain a multidimensional feature information group, and adding a label to each cluster result to mark whether the target corresponding to the cluster result is a multipath target; Constructing a feature data set based on a large number of labeled multidimensional feature information groups, and training the AdaBoost model based on the feature data set; During the training process, the AdaBoost algorithm is used to iteratively train multiple SVM base classifiers. In each iteration, the AdaBoost algorithm adjusts the sample weights according to the previous iteration results, so that the AdaBoost model pays more attention to samples with high error rates.

8. A multipath identification device, characterized in that: include: A signal processing module, used to acquire radar point cloud data in various scenes frame by frame, and use the radar point cloud data as original point cloud data to be identified, wherein the various scenes include at least one of a tunnel scene, a congested scene, an open scene, a straight scene, a curved road scene, and an iron guardrail scene; A clustering module, used for clustering the original point cloud data based on a density clustering algorithm to obtain cluster results, and associating the cluster results of the current frame with the cluster results of the previous frame, performing an association operation on the cluster results of all frames to obtain a cluster queue; A feature extraction module, configured to extract multi-dimensional features from each cluster result of the current frame according to a preset algorithm and combine the multi-dimensional feature units corresponding to all the cluster results of the current frame as a unit to obtain a multi-dimensional feature information group; A data preprocessing module, used for inputting the information of the multidimensional feature information group of the current frame into a pre-trained AdaBoost model, and outputting the multipath judgment results of each cluster result; The multipath detection module is used to determine that the target corresponding to the cluster result is a multipath target if the number of times that the cluster result in the cluster queue is marked as multipath in N consecutive frames exceeds a first preset threshold.

9. An electronic device, characterized in that: include: At least one processor, at least one memory and computer program instructions stored in the memory, which implement the method according to any one of claims 1 to 7 when the computer program instructions are executed by the processor.

10. A storage medium having computer program instructions stored thereon, characterized in that: When the computer program instructions are executed by a processor, the method according to any one of claims 1 to 7 is implemented.

Images