Millimeter wave radar ghosting suppression method and model based on machine learning

Abstract

The invention discloses a machine learning-based millimeter-wave radar ghosting suppression method and model, and belongs to the technical field of automatic target tracking. Data characteristic parameters closely related to radar characteristics and a subsequent target tracking algorithm are selected, and the characteristics include moving point cloud attributes directly influencing target generation and moving point cloud attributes directly influencing target generation; the model of the invention can identify and suppress ghosts more quickly and accurately, can be suitable for a specific scene by tightening or adding some constraint conditions, and also covers the environment information around the target although the target is not directly formed. And the method can be applied to a target tracking task in a composite complex environment by relaxing a constraint condition, so that the anti-interference capability of the system is improved, and a decision strategy for finally judging a ghosting and a real target by a tracking algorithm in different scenes is improved.

Description

Technical Field

[0001] The present invention relates to the technical field of automatic target tracking, and in particular to a millimeter wave radar ghost suppression method and model based on machine learning. Background Art

[0002] Millimeter-wave radar sensors play a vital role in advanced driver assistance systems (ADAS) and autonomous vehicles (AV) due to their excellent penetration and robustness to adverse weather conditions. These systems rely on radar sensors to detect other vehicles, pedestrians, obstacles, etc. in the surrounding environment to achieve advanced functions such as collision warning, adaptive cruise control, and automatic parking. However, millimeter-wave radar systems face some challenges in target tracking and environmental perception. One of them is the so-called "ghosting" phenomenon, which refers to false signals caused by factors such as multipath propagation, nonlinear effects, or hardware limitations in radar signal processing. The presence of ghosts seriously interferes with the radar system's detection and tracking of real targets, reducing the reliability and safety of the system.

[0003] The traditional ghost suppression methods currently have the following problems:

[0004] Although the use of radar point cloud to fit the drivable area boundary detection and road edge detection is effective to a certain extent, it lacks sufficient flexibility and adaptability to cope with the changing real-world traffic environment.

[0005] Suppression through the physical characteristics of ghost targets or through statistical analysis of features has limitations in accurately identifying and suppressing ghosts due to the complexity and variability of radar signals.

[0006] Based on this, the present invention designs a millimeter-wave radar ghost suppression method and model based on machine learning to solve the above problems. Summary of the invention

[0007] In view of the above-mentioned shortcomings of the prior art, the present invention provides a millimeter-wave radar ghost suppression method and model based on machine learning.

[0008] To achieve the above objectives, the present invention is implemented through the following technical solutions:

[0009] A millimeter wave radar ghost suppression method based on machine learning includes the following steps:

[0010] Step 1: Collect and preprocess the data of millimeter-wave radar target tracking under various environmental conditions, retrieve the data sequence corresponding to the target ID of each cycle in the HDF5 sequence file, perform feature processing on the data by retrieving the sequence name and target ID, and then extract the feature parameters that effectively distinguish the target signal and the interference signal, and then screen the feature parameters, and select the feature parameters with higher scores as the main input basis of the data set;

[0011] Step 2: Randomly divide the data set into training set and test set;

[0012] Step 3: In the submodule of the automatic parameter adjustment module, define the feature parameter search space and use grid search to find the optimal feature parameter combination: use the weighted combination of the Gini impurity criterion and the information entropy criterion to obtain a comprehensive evaluation index α. By adjusting α, you can make a trade-off between the Gini impurity criterion and the information entropy criterion, and then set the evaluation criteria for dividing the subtree, the number range of the base evaluator, the range of the minimum number of samples for the leaf node, and the range of the minimum number of samples required for the internal node to be re-divided. By setting these parameters reasonably, the training and prediction process of the model can be controlled;

[0013] Step 4: Build a classifier model in the automatic parameter adjustment module, use the random forest algorithm to build multiple decision trees, and then use the majority voting principle to determine the results of the integrated decision tree for training;

[0014] Step 5: After the decision tree is trained, the test set samples are input into the classifier model to calculate and compare the recall rate through the confusion matrix and ROC curve, and the AUC data is calculated to evaluate the performance of the classifier model. Then, the Monte Carlo method is used to build an automatic parameter adjustment framework. With the help of the computer's computing power, the above steps 3 and 4 are directly simulated. The feature parameters of the classifier are constructed by continuous sampling to find the expected recall rate and specificity, and the feature parameters with the most balanced recall rate and specificity are selected to gradually approach the best result. Finally, the best model feature parameters are further screened.

[0015] Step 6: Convert the optimal model feature parameters into a .h file. The generated .h file is represented by a binary tree data structure.

[0016] Step 7: In the radar-based object detection and tracking and environment perception modules, the ghost suppression model is integrated into the target tracking submodule. The target tracking submodule inputs the feature parameters into the target update module for calculation. After reading the feature parameters calculated in the target update module, the .h file is called to classify the target into a ghost target or a valid target. Each binary tree is recursively traversed, and a decision is made based on the feature parameters and threshold of the current node. The classified results are post-processed and a stable output result is obtained after counting with a counter.

[0017] Furthermore, in step 1, the data preprocessing process: the collected radar target tracking data is compared with the visual truth value, and annotated to distinguish between ghosts and real targets;

[0018] If the visual truth matches the target output of the tracking model, the is_val id label is set to 1;

[0019] If the visual truth does not match the target output of the tracking model, the is_val id label is set to 0;

[0020]

[0021] Where Z: target to be labeled, X: visual truth target, Y: model output target;

[0022] None: No matching target exists; Otherwise: X i ∩Y j ≠ the complement of None.

[0023] Furthermore, in step one, the characteristic parameters include the maximum radar cross-section area of ​​the i point cloud clusters associated with the target in the current cycle, the average of the maximum radar cross-section areas of the point cloud clusters associated with the target from the creation cycle m to the current maintenance cycle n, the maximum radar cross-section area of ​​the point cloud clusters associated with the target from the creation cycle m to the current maintenance cycle n, the maximum SNR of the point cloud cluster associated with the target, the ratio of the number and area of ​​static targets between the sensor installation position and the tracked target, and the target motion trajectory characteristics.

[0024] Furthermore, the calculation formula for the maximum radar cross-section area among the i point cloud clusters associated with the current periodic target is:

[0025] object_track.max_rcs=max(associated_det0.rcs,associated_det1.rcs

[0026] ,,,associated_det i-1 .rcs,associated_det i .rcs)

[0027] associated_det i : The i-th point cloud associated with the target;

[0028] The calculation formula for the mean value of the maximum radar cross-section in the point cloud group associated with the target from the creation cycle m to the current maintenance cycle n is:

[0029]

[0030] The formula for calculating the maximum radar cross-section of the point cloud group associated with the target from the creation cycle m to the current maintenance cycle n is:

[0031] object_track.max_max_rcs=max(object_track m .max_rcs,object_track m+ 1.max_rcs

[0032] ,,,object_track m+2 .max_rcs,object_track n-1 .max_rcs,object_track n .max_rcs)

[0033] The calculation formula for the ratio of the number of static targets to the area between the sensor installation position and the tracking target is:

[0034]

[0035] area_x=object_track.pos_x-sensor.pos_x

[0036] area_y=object_track.pos_y-sensor.pos_y

[0037] Among them, object_track.pos_x: the x coordinate of the target position in the coordinate system with the center of the vehicle's rear axle as the origin;

[0038] object_track.pos_y: the y coordinate of the target position in the coordinate system with the center of the vehicle's rear axle as the origin;

[0039] sensor.pos_x: x coordinate of the radar installation position;

[0040] sensor.pos_y: y coordinate of the radar installation position;

[0041] area_x: the difference in the x direction between the sensor installation position and the tracking target;

[0042] area_y: the difference in y direction between the sensor installation position and the tracking target;

[0043] The target motion trajectory characteristics are calculated by calculating the first-order derivative and second-order derivative of the target position.

[0044] Furthermore, in step three:

[0045] When α is 0, the information entropy criterion is preferred as the node division criterion;

[0046] When α is 1, the Gini impurity criterion is used as the node partitioning criterion.

[0047] Furthermore, the calculation method of node partitioning is:

[0048] criterion=α*Gini+(1-α)*Entropy

[0049] Gini=∑p(x)(1-p(x))

[0050] Entropy=-∑p(x)log2p(x)

[0051] Among them, criterion is the calculation standard for node division, Gini is the Gini coefficient, Entropy is the information entropy coefficient, and p(x) is the probability of category x.

[0052] Furthermore, in step six, the data structure of the binary tree includes a TreesNode structure and a Trees structure.

[0053] Furthermore, in step 6, the TreesNode structure is as follows:

[0054] typedef struct{

[0055] s8_t feature;

[0056] f32_t value;

[0057] s16_t left;

[0058] s16_t right;

[0059] }TreesNode;

[0060] The TreesNode structure is used to represent a node of a binary tree;

[0061] Feature: an 8-bit signed integer type (s8_t), indicating the feature number checked by the current node;

[0062] Value: a 32-bit floating point number type (f32_t), indicating the threshold of the current node;

[0063] Left: A 16-bit signed integer type (s16_t), which indicates the index of the left child of the current node in the nodes array; if it is a negative number, it means that this node is a leaf node and has no further child nodes;

[0064] Right: A 16-bit signed integer type (s16_t), which represents the index of the right child of the current node in the nodes array; if it is a negative number, it means that this node is a leaf node and has no further child nodes.

[0065] Furthermore, the Trees structure is as follows:

[0066] typedef struct{

[0067] s32_t n_nodes;

[0068] TreesNode* nodes;

[0069] s32_t n_trees;

[0070] s32_t*tree_roots;

[0071] }Trees;

[0072] The Trees structure is used to represent a forest consisting of multiple binary trees;

[0073] n_nodes: a 32-bit signed integer type (s32_t), representing the total number of nodes in all binary trees in the forest;

[0074] nodes: a pointer to an array of TreesNode structures, which stores the node information of all binary trees;

[0075] n_trees: a 32-bit signed integer type (s32_t), representing the number of binary trees in the forest;

[0076] tree_roots: A pointer to an array of 32-bit signed integer types (s32_t) that stores the index of the root node of each binary tree in the nodes array.

[0077] In order to better achieve the purpose of the present invention, the present invention also provides a ghost suppression model constructed by the above method.

[0078] Compared with the prior art, the present invention has the following beneficial effects:

[0079] 1. The present invention selects data feature parameters that are closely related to radar characteristics and subsequent target tracking algorithms. These data feature parameters include not only motion point cloud attributes that directly affect target generation, but also environmental information surrounding the target that does not directly form the target.

[0080] 2. The model generated by the present invention can be implemented by a binary tree, has strong interpretability, is easy to deploy in the target tracking submodule, and occupies little memory.

[0081] 3. Due to the existence of steps one and two, that is, the data feature parameters and restrictions are transparent to developers, the code is highly interpretable. When there is a large deviation between the output results and expectations, developers can quickly locate the problem and modify the code, greatly reducing the number and time of data training during the model algorithm tuning process.

[0082] 4. The model of the present invention can identify and suppress ghosts more quickly and accurately. It can be applied to a specific scenario by tightening or adding some constraints, and can also be applied to target tracking tasks in complex environments by relaxing constraints. It improves the system's anti-interference ability and improves the decision-making strategy of the tracking algorithm in different scenarios to finally determine ghosts and real targets.

[0083] 5. The present invention uses automated parameter adjustment design, which can reduce blind parameter adjustment and output the optimal model quickly and effectively. BRIEF DESCRIPTION OF THE DRAWINGS

[0084] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings required for use in the embodiments or the prior art descriptions are briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention, and for ordinary technicians in this field, other drawings can be obtained based on these drawings without creative work.

[0085] Figure 1 It is a flow chart of a millimeter wave radar ghost suppression method based on machine learning of the present invention;

[0086] Figure 2 It is a confusion matrix model evaluation diagram of the present invention;

[0087] Figure 3 This is a simulation diagram of the effect of machine learning on suppressing ghost targets in a tunnel scene in the present invention;

[0088] Figure 4 This is a simulation diagram of the effect of suppressing ghost targets using traditional suppression methods. DETAILED DESCRIPTION

[0089] In order to make the purpose, technical solution and advantages of the embodiments of the present invention clearer, the technical solution in the embodiments of the present invention will be clearly and completely described below in conjunction with the drawings in the embodiments of the present invention. Obviously, the described embodiments are part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0090] Embodiment 1

[0091] In some embodiments, please refer to the attached instructions. Figure 1 , a millimeter wave radar ghost suppression method based on machine learning, comprising the following steps:

[0092] Step 1: Collect and pre-process radar target tracking data under various environmental conditions: vehicles are stationary and moving in tunnels, elevated roads, and fences;

[0093] Data preprocessing process: compare the collected radar target tracking data with the visual truth value, and mark and distinguish ghosts from real targets;

[0094] If the visual truth matches the target output of the tracking model, the is_val id label is set to 1, otherwise it is set to 0;

[0095]

[0096] Z: target to be labeled; X: visual truth target; Y: model output target;

[0097] None: No matching target exists; Otherwise: X i ∩Y j ≠ the complement of None.

[0098] Retrieve the data sequence corresponding to the target ID of each cycle in the HDF5 sequence file, perform feature processing on the data by retrieving the sequence name and target ID, and then extract the feature parameters that effectively distinguish the target signal from the interference signal;

[0099] For each feature parameter, set the interval range in segments and maintain a corresponding counter. Then use the counters relative to the feature parameters to score and screen the features to adaptively learn and update the recognition ability of the features. The initial value of the counter is 0. When a feature parameter is in this interval, the corresponding counter is increased by 1. The counter value of each feature parameter is calculated, and the frequency ratio of each feature parameter in the valid target and ghost target set is calculated as the score of the feature. Finally, the feature parameter with a higher score is selected as the main input basis of the data set for identifying the target.

[0100] The characteristic parameters include:

[0101] Calculate the maximum RCS (Radar Cross Section) in the point cloud group associated with the target in the current period, that is, the maximum radar cross-section and the average value in the point cloud group associated with the target in the current period from creation to maintenance; this dynamic time series analysis can better capture the motion characteristics of the target and help distinguish between real targets that persist and short-lived interference signals.

[0102] RCS is an indicator that measures the target's ability to reflect radar signals in the direction of radar reception. The larger the RCS, the easier it is to detect the object.

[0103] The calculation formula for the largest radar cross-section among the i point cloud clusters associated with the current period target is:

[0104] object_track.max_rcs=max(associated_det0.rcs,associated_det1.rcs

[0105] ,,,associated_det i-1 .rcs,associated_det i .rcs)

[0106] associated_det i : The i-th point cloud associated with the label;

[0107] The average maximum radar cross-section of the point cloud cluster associated with the target from creation cycle m to the current maintenance cycle n is:

[0108]

[0109] The formula for calculating the maximum radar cross-section of the point cloud group associated with the target from the creation cycle m to the current maintenance cycle n is:

[0110] object_track.max_max_rcs=max(object_track m .max_rcs,object_track m+ 1.max_rcs

[0111] ,,,object_track m+2 .max_rcs,object_track n-1 .max_rcs,object_track n .max_rcs)

[0112] The maximum SNR (signal-to-noise ratio) in the point cloud cluster associated with the target. The signal-to-noise ratio indicates the transmission quality of the signal under noise interference. Generally, the larger the signal-to-noise ratio, the better the signal quality.

[0113] The presence of a reflective mirror is determined by calculating the ratio of the number and area of ​​static targets between the sensor installation position and the tracking target;

[0114] The calculation formula for the ratio of the number of static targets to the area between the sensor installation position and the tracking target is:

[0115]

[0116] area_x=object_track.pos_x-sensor.pos_x

[0117] area_y=object_track.pos_y-sensor.pos_y

[0118] Among them, object_track.pos_x: the x coordinate of the target position in the coordinate system with the center of the vehicle's rear axle as the origin;

[0119] object_track.pos_y: the y coordinate of the target position in the coordinate system with the center of the vehicle's rear axle as the origin;

[0120] sensor.pos_x: x coordinate of the radar installation position;

[0121] sensor.pos_y: y coordinate of the radar installation position;

[0122] area_x: the difference in the x direction between the sensor installation position and the tracking target;

[0123] area_y: The difference in y direction between the sensor installation position and the tracking target.

[0124] Calculate the first-order derivative (dx / dt, dy / dt) and second-order derivative (d 2 x / dt2,d 2 y / dt2) to obtain the target motion trajectory characteristics to describe the smoothness of the target motion trajectory. The real target usually has a relatively smooth and continuous motion trajectory, while the trajectory of the ghost is relatively messy and discontinuous.

[0125] In addition to the characteristic parameters proposed above, the characteristic parameters also include: target lateral velocity (velX_mps);

[0126] Target longitudinal velocity (velY_mps);

[0127] First order derivative of target lateral velocity (first_order_derivat ive_velX_mps);

[0128] The second-order derivative of the target lateral velocity (second_order_derivat ive_velX_mps);

[0129] The first-order derivative of the target longitudinal velocity (first_order_derivat ive_velY_mps), etc.

[0130] Step 2: Randomly divide the data set into training set and test set; the division method is: combine the ID (identity document) and time stamp (timestamp signal) of the target in the data set as a unique identifier, then divide the target according to a certain ratio (80% training set, 20% test set), and finally extract the rest of its related information through this unique identifier.

[0131] Step 3: Before entering the automatic parameter tuning module, fix the random_state parameter to ensure the repeatability of the results. The random_state parameter is used to control the randomness in the random forest. By fixing the random_state parameter, the randomness of these random processes can be controlled so that the random forest model generated each time is the same;

[0132] The value of the class_weight (sample class weight) parameter is determined according to the ratio of the number of valid targets to ghost targets in the training set samples. By setting the class_weight parameter, the weight of the minority class can be increased, so that the model pays more attention to the minority class, thereby improving the performance of the overall model;

[0133] Since the number of samples of different categories in the data set varies greatly, this may cause the model to favor the majority class and ignore the minority class during training. The class_weight parameter can be used to effectively deal with the class imbalance problem;

[0134] Then, in the submodule of the automatic parameter adjustment module, the feature parameter search space is defined, and the grid search is used to find the optimal feature parameter combination: a weighted combination of the Gini impurity criterion and the information entropy criterion is used to obtain a comprehensive evaluation index α. By adjusting α, a trade-off can be made between the two aspects.

[0135] The Gini degree criterion focuses on local purity, while the information entropy criterion focuses on global uncertainty. Both are indicators that reflect data purity, but they have different emphases.

[0136] In the initial stage, the data has not been fully divided and is still relatively messy. At this time, α approaches 0, which means that information entropy is more inclined to be used as the division criterion. As the decision tree continues to grow, the data has been well divided at each node. At this time, α approaches 1, which means that it is more appropriate to use the Gini degree criterion as the division criterion, because the Gini degree criterion pays more attention to local purity and can better fine-tune and optimize the division of each node. The calculation formula for node division is:

[0137] criterion=α*Gini+(1-α)*Entropy

[0138] Gini=∑p(x)(1-p(x))

[0139] Entropy=-∑p(x)log2p(x)

[0140] Among them, criterion is the calculation standard for node division, Gini is the Gini coefficient, Entropy is the information entropy coefficient, and p(x) is the probability of category x.

[0141] Set the evaluation criteria for subtree partitioning (criterion), the range of the number of base estimators (n_estimators), the range of the minimum number of samples for leaf nodes (min_samples_leaf), and the range of the minimum number of samples required for internal node re-division (min_samples_split);

[0142] The criterion parameter is used to specify the splitting criterion of the decision tree;

[0143] The n_estimators parameter is used to specify the number of decision trees in the random forest;

[0144] The min_samples_split parameter is used to specify the minimum number of samples required for a node to split. If a node contains fewer samples than this value, the node will not be split.

[0145] The min_samples_leaf parameter is used to specify the minimum number of samples required for a leaf node. If a leaf node contains fewer samples than this value, the leaf node will be merged.

[0146] By setting these parameters reasonably, the model training and prediction process can be controlled and the model performance and generalization ability can be improved.

[0147] Step 4: Build a classifier model in the automatic parameter adjustment module, use the random forest algorithm to build multiple decision trees, and then use the majority voting principle to determine the results of the integrated decision tree for training; each tree is built based on a random subset of the training set, and each tree grows independently during the training process until it cannot split further or reaches the preset stopping condition;

[0148] At the same time, pruning is used to prevent overfitting, where the decision tree accurately classifies the training data but has a poor classification effect on unknown test data.

[0149] Step 5: After the decision tree is trained, the test set samples are input into the classifier model through the confusion matrix and ROC (Receiver Operating Characteristic Curve) curve, and the AUC (Area Under Curve) is calculated to evaluate the performance of the classifier model. The Monte Carlo method is then used to build an automatic parameter adjustment framework. The above steps 3 and 4 are directly simulated with the help of the computer's computing power. The classifier parameters are continuously sampled to find the expected recall rate and specificity, and the parameters with the most balanced recall rate and specificity are selected to gradually approach the best result. Finally, the best model parameters are further screened out.

[0150] For each feature parameter combination, the model is trained using the training set and its performance is evaluated on the test set. When evaluating the model, the recall rate is used as the first evaluation criterion to minimize the situation where valid targets are mistakenly suppressed as ghost images. Based on the model evaluation results and the AUC value, the optimal feature parameter combination can be selected to achieve the best balance between recall rate and specificity.

[0151]

[0152] TP (True Positive): True positive, that is, the number of samples correctly predicted by the model as positive;

[0153] FN (False Negative): False negative, that is, the number of samples that the model incorrectly predicts as negative;

[0154]

[0155] TN (True Negative): True negative, that is, the number of samples correctly predicted by the model as negative;

[0156] FP (False Positive): False positive, that is, the number of samples that the model incorrectly predicts as positive.

[0157] Step 6: Convert the optimal model feature parameters into a .h file for easy integration and deployment in embedded systems or other hardware platforms, and integrate it into the target tracking submodule to perform classification judgments on each target in real time to suppress ghost interference and improve the accuracy and reliability of target tracking. The target dynamic information calculated by the tracking module is directly fed back to the classification model. At the same time, the classification results are directly fed back to the tracking module and combined with the calculated multiple attributes to promote the optimization of the overall classification performance.

[0158] The generated .h file is represented by a binary tree data structure, which includes a TreesNode structure and a Trees structure. Compared with using the traditional if-else (if conditional statement) to judge the value of each node, it can save about 80% of the memory space and also improve the calculation speed.

[0159] This binary tree data structure can efficiently represent and deploy machine learning models, and is very suitable for application scenarios in embedded systems and other hardware platforms.

[0160] The data structure of a binary tree is as follows:

[0161] The TreesNode structure is as follows:

[0162] typedef struct{

[0163] s8_t feature;

[0164] f32_t value;

[0165] s16_t left;

[0166] s16_t right;

[0167] }TreesNode;

[0168] The TreesNode structure is used to represent a node of a binary tree;

[0169] Feature: an 8-bit signed integer type (s8_t), indicating the feature number checked by the current node;

[0170] Value: a 32-bit floating point number type (f32_t), indicating the threshold of the current node;

[0171] Left: A 16-bit signed integer type (s16_t), which indicates the index of the left child of the current node in the nodes array; if it is a negative number, it means that this node is a leaf node and has no further child nodes;

[0172] Right: A 16-bit signed integer type (s16_t), which represents the index of the right child of the current node in the nodes array; if it is a negative number, it means that this node is a leaf node and has no further child nodes.

[0173] The Trees structure is as follows:

[0174] typedef struct{

[0175] s32_t n_nodes;

[0176] TreesNode* nodes;

[0177] s32_t n_trees;

[0178] s32_t*tree_roots;

[0179] }Trees;

[0180] The Trees structure is used to represent a forest consisting of multiple binary trees;

[0181] n_nodes: a 32-bit signed integer type (s32_t), representing the total number of nodes in all binary trees in the forest;

[0182] nodes: a pointer to an array of TreesNode structures, which stores the node information of all binary trees;

[0183] n_trees: a 32-bit signed integer type (s32_t), representing the number of binary trees in the forest;

[0184] tree_roots: A pointer to an array of 32-bit signed integer types (s32_t) that stores the index of the root node of each binary tree in the nodes array.

[0185] Step 7: In the radar-based object detection and tracking and environment perception modules, the generated ghost suppression model is integrated into the target update submodule. The target tracking submodule inputs the feature parameters into the target update module for calculation. After reading the feature parameters calculated in the target update module, the .h file is called to classify the target into a ghost target or a valid target. Each binary tree is recursively traversed, and a decision is made based on the features and thresholds of the current node. The classified results are post-processed, and a stable output result is obtained after counting with a counter.

[0186] The output result is saved in is_ghost_by_machine_learning (whether the target is determined to be a ghost by machine learning). Later, is_ghost_by_machine_learning can be combined with other attributes of the target, such as is_crossing (whether the target is crossing), is_created_near_fov (whether the target is created at the edge of FOV), etc., to assign is_ghost (whether the target is a ghost) values ​​of targets in different complex environments.

[0187] In order to better achieve the purpose of the present invention, the present invention also provides a ghost suppression model constructed by the above method.

[0188] Embodiment 2

[0189] In some embodiments, Figure 2 As shown, the confusion matrix is ​​used to evaluate the final generated model to calculate the F1_Score (the harmonic mean of precision and recall to reflect the overall performance of the model in a balanced way). The calculation method of F1_Score is as follows:

[0190]

[0191]

[0192] Depend on Figure 2 From the data, we can conclude that the value of F1_Score is 0.984.

[0193] Embodiment 3

[0194] In some embodiments, Figure 3-4 As shown, Figure 3 This is a simulation diagram of the effect of traditional suppression methods on ghost targets. Figure 4 The simulation diagram of the effect of suppressing ghost targets in a tunnel scene using machine learning. It can be seen that the present invention solves the limitation that free space cannot suppress ghost targets, and successfully suppresses false targets generated by moving noise caused by angle measurement errors.

[0195] The above embodiments are only used to illustrate the technical solutions of the present invention, rather than to limit the same. Although the present invention has been described in detail with reference to the aforementioned embodiments, those skilled in the art should understand that the technical solutions described in the aforementioned embodiments may still be modified, or some of the technical features may be replaced by equivalents. Such modifications or replacements will not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of the present invention.

Claims

1. A millimeter wave radar ghost suppression method based on machine learning, characterized in that: The following steps are involved: Step 1: Collect and preprocess the data of millimeter-wave radar target tracking under various environmental conditions, retrieve the data sequence corresponding to the target ID of each cycle in the HDF5 sequence file, perform feature processing on the data by retrieving the sequence name and target ID, and then extract the feature parameters that effectively distinguish the target signal and the interference signal, and then screen the feature parameters, and select the feature parameters with higher scores as the main input basis of the data set; Step 2: Randomly divide the data set into training set and test set; Step 3: In the submodule of the automatic parameter adjustment module, define the feature parameter search space and use grid search to find the optimal feature parameter combination: use the weighted combination of the Gini impurity criterion and the information entropy criterion to obtain a comprehensive evaluation index α. By adjusting α, you can make a trade-off between the Gini impurity criterion and the information entropy criterion, and then set the evaluation criteria for dividing the subtree, the number range of the base evaluator, the range of the minimum number of samples for the leaf node, and the range of the minimum number of samples required for the internal node to be re-divided; by setting these parameters reasonably, you can control the training and prediction process of the model; Step 4: Build a classifier model in the automatic parameter adjustment module, use the random forest algorithm to build multiple decision trees, and then use the majority voting principle to determine the results of the integrated decision tree for training; Step 5: After the decision tree is trained, the test set samples are input into the classifier model to calculate and compare the recall rate through the confusion matrix and ROC curve, and the AUC data is calculated to evaluate the performance of the classifier model. Then, the Monte Carlo method is used to build an automatic parameter adjustment framework. With the help of the computer's computing power, the above steps 3 and 4 are directly simulated. The feature parameters of the classifier are constructed by continuous sampling to find the expected recall rate and specificity, and the feature parameters with the most balanced recall rate and specificity are selected to gradually approach the best result. Finally, the best model feature parameters are further screened. Step 6: Convert the optimal model feature parameters into a .h file. The generated .h file is represented by a binary tree data structure. Step 7: In the radar-based object detection and tracking and environment perception modules, the ghost suppression model is integrated into the target tracking submodule. The target tracking submodule inputs the feature parameters into the target update module for calculation. After reading the feature parameters calculated in the target update module, the .h file is called to classify the target into a ghost target or a valid target. Each binary tree is recursively traversed, and a decision is made based on the feature parameters and threshold of the current node. The classified results are post-processed and a stable output result is obtained after counting with a counter.

2. The method for suppressing ghost images of millimeter wave radar based on machine learning according to claim 1, characterized in that: In step 1, the data preprocessing process: the collected radar target tracking data is compared with the visual truth value, and annotated to distinguish ghosts from real targets; If the visual truth matches the target output of the tracking model, the is_valid label is set to 1; If the visual truth does not match the target output of the tracking model, the is_valid label is set to 0; Where Z: target to be labeled, X: visual truth target, Y: model output target; None: No matching target exists; Otherwise: X i ∩Y j ≠ the complement of None.

3. The method for suppressing ghost images of millimeter wave radar based on machine learning according to claim 2, characterized in that: In step one, the characteristic parameters include the maximum radar cross-section area of ​​the i point cloud clusters associated with the target in the current cycle, the average of the maximum radar cross-section areas of the point cloud clusters associated with the target from the creation cycle m to the current maintenance cycle n, the maximum radar cross-section area of ​​the point cloud clusters associated with the target from the creation cycle m to the current maintenance cycle n, the maximum SNR of the point cloud cluster associated with the target, the ratio of the number and area of ​​static targets between the sensor installation position and the tracked target, and the target motion trajectory characteristics.

4. The method for suppressing ghost images of millimeter wave radar based on machine learning according to claim 3 is characterized in that: The calculation formula for the largest radar cross-section among the i point cloud clusters associated with the current period target is: object_track.max_rcs=max(associated_det0.rcs,associated_det1.rcs ,,,associated_it i-1 .rcs,associated_det i .rcs) associated_det i : The i-th point cloud associated with the target; The calculation formula for the mean value of the maximum radar cross-section in the point cloud group associated with the target from the creation cycle m to the current maintenance cycle n is: The formula for calculating the maximum radar cross-section of the point cloud group associated with the target from the creation cycle m to the current maintenance cycle n is: object_track.max_max_rcs=max(object_track m .max_rcs,object_track m+1 .max_rcs ,,,object_track m+2 .max_rcs,object_track n-1 .max_rcs,object_track n .max_rcs) The calculation formula for the ratio of the number of static targets to the area between the sensor installation position and the tracking target is: area_x=object_track.pos_x-sensor.pos_x area_y=object_track.pos_y-sensor.pos_y Among them, object_track.pos_x: the x coordinate of the target position in the coordinate system with the center of the vehicle's rear axle as the origin; object_track.pos_y: the y coordinate of the target position in the coordinate system with the center of the vehicle's rear axle as the origin; sensor.pos_x: the x coordinate of the radar installation position; sensor.pos_y: y coordinate of the radar installation position; area_x: the difference in the x direction between the sensor installation position and the tracking target; area_y: the difference in y direction between the sensor installation position and the tracking target; The target motion trajectory characteristics are calculated by calculating the first-order derivative and second-order derivative of the target position.

5. The method for suppressing ghost images of millimeter wave radar based on machine learning according to claim 4, characterized in that: In step three: When α is 0, the information entropy criterion is preferred as the node division criterion; When α is 1, the Gini impurity criterion is used as the node partitioning criterion.

6. The method for suppressing ghost images of millimeter wave radar based on machine learning according to claim 5, characterized in that: Calculation method for node division: criterion=α*Gini+(1-α)*Entropy Gini=∑p(x)(1-p(x)) Entropy=-∑p(x)log2p(x) Among them, criterion is the calculation standard for node division, Gini is the Gini coefficient, Entropy is the information entropy coefficient, and p(x) is the probability of category x.

7. The method for suppressing ghost images of millimeter wave radar based on machine learning according to claim 6, characterized in that: In step six, the data structure of the binary tree includes the TreesNode structure and the Trees structure.

8. The method for suppressing ghost images of millimeter wave radar based on machine learning according to claim 7, characterized in that: The TreesNode structure is as follows: typedef struct{ s8_t feature; f32_t value; s16_t left; s16_t right; }TreesNode.

9. The method for suppressing ghost images of millimeter wave radar based on machine learning according to claim 8, characterized in that: The Trees structure is as follows: typedef struct{ s32_t n_nodes; TreesNode* nodes; s32_t n_trees; s32_t*tree_roots; }Trees.

10. A ghost suppression model constructed using the machine learning-based millimeter wave radar ghost suppression method according to claim 9.

Images