Label-based vehicle-mounted millimeter-wave radar multi-target detection method and system

Abstract

The present invention relates to the field of target detection technology, and provides a label-based vehicle-mounted millimeter-wave radar multi-target detection method and system. The method includes: obtaining a multi-target predicted state set based on a multi-target state set prediction at a first moment; updating the weights of particles based on a measurement set obtained by detection of a vehicle-mounted millimeter-wave radar at a second moment; adjusting the labels of particles based on the updated weights; clustering particles in the multi-target predicted state set at the second moment based on the labels, and weighted averaging the clustering results based on the weights to obtain the multi-target detection results at the second moment. The present invention detects and identifies the states of multiple targets through a label-based particle filter, thereby avoiding unnecessary association of detection targets; at the same time, the technical means of updating weights through measurement and adjusting labels through weights can further enhance the accuracy of multi-target recognition results.

Description

Technical Field

[0001] The present invention relates to the field of target detection technology, and in particular to a label-based vehicle-mounted millimeter-wave radar multi-target detection method and system. Background Art

[0002] Autonomous driving, also known as unmanned driving, computer driving or wheeled mobile robot, is a cutting-edge technology that relies on computers and artificial intelligence technology to complete complete, safe and effective driving without human control.

[0003] In the 21st century, due to the continuous increase in car users, road traffic is facing more and more serious problems such as congestion and safety accidents. With the support of vehicle networking technology and artificial intelligence technology, autonomous driving technology can coordinate travel routes and planning time, thereby greatly improving travel efficiency and reducing energy consumption to a certain extent. Autonomous driving can also help avoid safety hazards such as drunk driving and fatigue driving, reduce driver errors, and improve safety. Therefore, autonomous driving has become a research and development focus of various countries in recent years.

[0004] As an automated vehicle, a self-driving car can sense its environment and navigate without human operation. As a feasible hardware for autonomous driving environment perception, the on-board millimeter-wave radar can collect point cloud data of obstacles during driving. Furthermore, based on the point cloud data, the state of obstacles, such as the position, speed, and size of multiple targets, can be analyzed.

[0005] In the scenario of autonomous driving or assisted driving, the target association task in multi-target detection is difficult and challenging. For the target association task, the existing technology usually first associates the target state and the target measurement, and then filters the combination with the largest association probability.

[0006] However, similar association methods cannot effectively overcome the problems of false detection and interference, and are prone to association errors, which in turn lead to target loss. In other words, the accuracy of target association in the existing technology is difficult to meet the requirements of autonomous driving or assisted driving scenarios, especially high-speed scenarios.

[0007] Therefore, how to provide a multi-target detection method with higher target association accuracy has become a technical problem that needs to be urgently solved in the industry. Summary of the invention

[0008] The present invention provides a label-based vehicle-mounted millimeter-wave radar multi-target detection method and system, which are used to solve the defect of low target association accuracy in the prior art and achieve more accurate multi-target recognition.

[0009] The present invention provides a label-based vehicle-mounted millimeter-wave radar multi-target detection method, comprising:

[0010] Get the multi-target state set at the first moment;

[0011] Predicting a response state set at a second moment according to the multi-objective state set at the first moment, and taking the union of the response state set at the second moment and the new state set as the multi-objective predicted state set;

[0012] The multi-target state set includes target state particles; the response state set includes response particles that survive or are derived from the target state particles; the new state set includes new particles that are sampled in the measurement area;

[0013] The target state particle includes a target state matrix, a label and a weight; the response particle includes a response state matrix, a label and a weight; the newborn particle includes a set newborn state matrix, a newborn label and a newborn weight;

[0014] According to the measurement set obtained by the detection of the vehicle-mounted millimeter-wave radar at the second moment, updating the weights of the particles in the multi-target prediction state set at the second moment;

[0015] Adjusting the labels of the particles in the multi-objective prediction state set at the second moment according to the updated weights of the particles in the multi-objective prediction state set at the second moment;

[0016] The particles in the multi-target prediction state set at the second moment are clustered according to the labels, and the clustering results are weighted averaged according to the weights to obtain the multi-target detection result at the second moment.

[0017] According to a label-based vehicle-mounted millimeter-wave radar multi-target detection method provided by the present invention, the step of adjusting the labels of particles in the multi-target predicted state set at the second moment according to the updated weights of the particles in the multi-target predicted state set at the second moment comprises:

[0018] According to the sum of the weights of the particles in the updated multi-objective prediction state set at the second moment, the estimated value of the number of targets at the second moment is obtained, and the estimated value of the number of targets at the second moment is judged:

[0019] Determine the difference N between the target quantity estimate at the second moment and the target quantity estimate at the first moment new is greater than zero, the label of the new target particle is adjusted to a value different from the label of the existing particle; the new target particle is a newly-born particle in the multi-target prediction state set at the second moment whose updated weight is greater than the first threshold, and the number of the new target particles is not greater than N new ;

[0020] Determine the difference N between the target quantity estimate at the second moment and the target quantity estimate at the first momentnew If it is equal to zero, the existing label is kept;

[0021] Determine the difference N between the target quantity estimate at the second moment and the target quantity estimate at the first moment new is less than zero, the label of the disappeared target particle is adjusted to the same value as the new label; the disappeared target particle is a particle with the same label whose updated weight is less than the second threshold in the multi-target prediction state set at the second moment.

[0022] According to a label-based vehicle-mounted millimeter-wave radar multi-target detection method provided by the present invention, the step of obtaining a multi-target state set at a first moment includes:

[0023] Excluding particles with new labels in the multi-target prediction state set at the first moment, and obtaining the set to be resampled at the first moment;

[0024] Based on the set to be resampled at the first moment, the multi-target state set at the first moment is resampled; in the multi-target state set at the first moment, the particle density in the set area is positively correlated with the particle weight in the set area in the set to be resampled at the first moment.

[0025] A label-based vehicle-mounted millimeter-wave radar multi-target detection method provided by the present invention also includes:

[0026] Determine a single target detection result of each target in the multi-target detection results according to the label;

[0027] Connecting the single target detection results of the same target at adjacent moments within the determination period to obtain the single target trajectory within the determination period;

[0028] Determining that the length of the single target trajectory is less than a third threshold, generating a false detection record;

[0029] If it is determined that there is a breakpoint in the single target trajectory, a missed detection record is generated.

[0030] According to a label-based vehicle-mounted millimeter-wave radar multi-target detection method provided by the present invention, the step of predicting a response state set at a second moment according to a multi-target state set at the first moment comprises:

[0031] According to the multi-target state set at the first moment, predict the particle state based on the uniform linear motion model to obtain an intermediate prediction result;

[0032] According to the a priori prediction probability error range, an error disturbance term is added to the intermediate prediction result to obtain a response state set at the second moment.

[0033] According to a label-based vehicle-mounted millimeter-wave radar multi-target detection method provided by the present invention, the step of updating the weights of particles in the multi-target prediction state set at the second moment according to the measurement set detected by the vehicle-mounted millimeter-wave radar at the second moment comprises:

[0034] Normalizing the measurement set obtained by the detection of the vehicle-mounted millimeter-wave radar at the second moment to obtain a measurement normalized set;

[0035] According to the measurement normalization set, the weights of the particles in the multi-objective prediction state set at the second moment are updated.

[0036] The present invention also provides a label-based vehicle-mounted millimeter-wave radar multi-target detection system, comprising:

[0037] An acquisition module, used to acquire a multi-target state set at a first moment;

[0038] A prediction module, configured to predict a response state set at a second moment according to the multi-objective state set at the first moment, and use a union of the response state set at the second moment and the new state set as the multi-objective prediction state set;

[0039] The multi-target state set includes target state particles; the response state set includes response particles that survive or are derived from the target state particles; the new state set includes new particles that are sampled in the measurement area;

[0040] The target state particle includes a target state matrix, a label and a weight; the response particle includes a response state matrix, a label and a weight; the newborn particle includes a set newborn state matrix, a newborn label and a newborn weight;

[0041] A weight module, used to update the weights of particles in the multi-target prediction state set at the second moment according to the measurement set detected by the vehicle-mounted millimeter-wave radar at the second moment;

[0042] A label module, used for adjusting the labels of the particles in the multi-objective prediction state set at the second moment according to the updated weights of the particles in the multi-objective prediction state set at the second moment;

[0043] The detection module is used to cluster the particles in the multi-target prediction state set at the second moment according to the labels, and perform weighted averaging on the clustering results according to the weights to obtain the multi-target detection result at the second moment.

[0044] The present invention also provides an electronic device, comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein when the processor executes the program, the steps of any of the above-described label-based vehicle-mounted millimeter-wave radar multi-target detection methods are implemented.

[0045] The present invention also provides a non-transitory computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the steps of any of the above-described label-based vehicle-mounted millimeter-wave radar multi-target detection methods.

[0046] The present invention also provides a computer program product, comprising a computer program, which, when executed by a processor, implements the steps of any of the above-mentioned label-based vehicle-mounted millimeter-wave radar multi-target detection methods.

[0047] The label-based vehicle-mounted millimeter-wave radar multi-target detection method and system provided by the present invention detects and identifies the states of multiple targets through a label-based particle filter, thereby avoiding unnecessary association of detection targets;

[0048] At the same time, the accuracy of multi-target recognition results can be further enhanced by measuring and updating weights and adjusting labels through weights. BRIEF DESCRIPTION OF THE DRAWINGS

[0049] In order to more clearly illustrate the technical solutions in the present invention or the prior art, the following briefly introduces the drawings required for use in the embodiments or the description of the prior art. Obviously, the drawings described below are some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative work.

[0050] Figure 1 It is a flow chart of the label-based vehicle-mounted millimeter-wave radar multi-target detection method provided by the present invention;

[0051] Figure 2 It is a structural schematic diagram of a label-based vehicle-mounted millimeter-wave radar multi-target detection device provided by the present invention;

[0052] Figure 3 It is a structural schematic diagram of the electronic device provided by the present invention.

[0053] Reference numerals:

[0054] 1: Acquisition module; 2: Prediction module; 3: Weight module;

[0055] 4: label module; 5: detection module; 310: processor;

[0056] 320: communication interface; 330: memory; 340: communication bus. DETAILED DESCRIPTION

[0057] In order to make the purpose, technical solution and advantages of the present invention clearer, the technical solution of the present invention will be clearly and completely described below in conjunction with the drawings 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.

[0058] Combine the following Figure 1 The present invention describes the label-based vehicle-mounted millimeter-wave radar multi-target detection method.

[0059] like Figure 1 As shown, an embodiment of the present invention provides a label-based vehicle-mounted millimeter-wave radar multi-target detection method, including:

[0060] Step 101, obtaining a multi-target state set at a first moment;

[0061] Step 103, predicting a response state set at a second moment according to the multi-objective state set at the first moment, and taking the union of the response state set at the second moment and the new state set as the multi-objective predicted state set;

[0062] The multi-target state set includes target state particles; the response state set includes response particles that survive or are derived from the target state particles; the new state set includes new particles that are sampled in the measurement area;

[0063] The target state particle includes a target state matrix, a label and a weight; the response particle includes a response state matrix, a label and a weight; the newborn particle includes a set newborn state matrix, a newborn label and a newborn weight;

[0064] Step 105, updating the weights of particles in the multi-target prediction state set at the second moment according to the measurement set detected by the vehicle-mounted millimeter-wave radar at the second moment;

[0065] Step 107, adjusting the labels of the particles in the multi-objective prediction state set at the second moment according to the updated weights of the particles in the multi-objective prediction state set at the second moment;

[0066] Step 109: cluster the particles in the multi-target prediction state set at the second moment according to the labels, and perform weighted averaging on the clustering results according to the weights to obtain the multi-target detection result at the second moment.

[0067] In a preferred embodiment, the first moment is earlier than the second moment, and the first moment and the second moment are adjacent moments related to the acquisition frequency of the vehicle-mounted millimeter-wave radar;

[0068] For example, if the acquisition frequency of the vehicle-mounted millimeter-wave radar is 10 Hz, that is, 10 times per second, and the moment of the 0th second is taken as the first moment, the second moment should be the moment of the 0.1th second.

[0069] It is worth noting that this embodiment can be understood as a repetitive unit in an iterative process. The method of this embodiment can obtain the multi-target detection result at the current moment through the multi-target detection result at the previous moment (or the multi-target prediction state set with updated weights and adjusted labels) and the measurement set at the current moment.

[0070] If the multi-target detection result at the previous moment (or the multi-target prediction state set with updated weights and adjusted labels) does not exist (for example, when initializing the method of this embodiment, or when multiple targets did not exist at the previous moment), the set initial multi-target state set can be used as the multi-target detection result at the previous moment;

[0071] An optional initial multi-objective state set includes initial particles uniformly set in the drivable area, and the probability of each initial particle is a set value (but the sum of the probabilities of the initial particles should be 1).

[0072] In a preferred embodiment, the step of obtaining a multi-target state set at a first moment includes:

[0073] Excluding particles with new labels in the multi-target prediction state set at the first moment, and obtaining the set to be resampled at the first moment;

[0074] Based on the set to be resampled at the first moment, the multi-target state set at the first moment is resampled; in the multi-target state set at the first moment, the particle density in the set area is positively correlated with the particle weight in the set area in the set to be resampled at the first moment.

[0075] In step 103, in the response state set, the response particle is used as a surviving or derived target state particle, and its label value and weight value remain unchanged; the measurement area refers to the area where the vehicle-mounted millimeter-wave radar has a measurement value at the first moment (that is, the measurement area is obtained by extending the set distance range through the measurement value position point of the vehicle-mounted millimeter-wave radar at the first moment);

[0076] In step 109, the label-based clustering operation may exclude invalid particles whose label values ​​are specified values ​​(eg, label values ​​are new labels).

[0077] The beneficial effects of this embodiment are:

[0078] The states of multiple targets are detected and identified through a label-based particle filter, thus avoiding unnecessary association of detected targets;

[0079] At the same time, the accuracy of multi-target recognition results can be further enhanced by measuring and updating weights and adjusting labels through weights.

[0080] According to the above embodiment, in this embodiment:

[0081] The step of adjusting the labels of the particles in the multi-objective prediction state set at the second moment according to the updated weights of the particles in the multi-objective prediction state set at the second moment comprises:

[0082] According to the sum of the weights of the particles in the updated multi-objective prediction state set at the second moment, the estimated value of the number of targets at the second moment is obtained, and the estimated value of the number of targets at the second moment is judged:

[0083] Determine the difference N between the target quantity estimate at the second moment and the target quantity estimate at the first moment new is greater than zero, the label of the new target particle is adjusted to a value different from the label of the existing particle; the new target particle is a newly-born particle in the multi-target prediction state set at the second moment whose updated weight is greater than the first threshold, and the number of the new target particles is not greater than N new ;

[0084] Determine the difference N between the target quantity estimate at the second moment and the target quantity estimate at the first moment new If it is equal to zero, the existing label is kept;

[0085] Determine the difference N between the target quantity estimate at the second moment and the target quantity estimate at the first moment new is less than zero, the label of the disappeared target particle is adjusted to the same value as the new label; the disappeared target particle is a particle with the same label whose updated weight is less than the second threshold in the multi-target prediction state set at the second moment.

[0086] The method for obtaining the estimated value of the target quantity at the second moment is as follows:

[0087] The total number of targets was calculated and an estimate of the number of each target was calculated based on particles with the same label.

[0088] The total number of targets can be estimated by summing the weights of the particle sets that fit the multi-target posterior PHD.

[0089] The weights of the box particles with the same label in the label set are used as the number estimate of a single target, so that the number estimate of each target can be obtained. In the case of missed detection, it can be distinguished which target has not been detected.

[0090] In a preferred embodiment, the method for obtaining new target particles is:

[0091] Search for particles with new labels, sort the particles with weights greater than the first threshold according to their weights, and sort the first N particles with larger weights. new The particle is taken as the new target particle, and the label of the new target particle is adjusted to a value different from the label of the existing particle;

[0092] If among the particles with new labels, the number of particles with weights greater than the first threshold is less than N new If there are all particles with new labels and weights greater than the first threshold, they are taken as new target particles.

[0093] The beneficial effects of this embodiment are:

[0094] According to the target number estimation result, the change in the number of targets is determined. When the number of targets increases, particles with higher probability are selected as new targets for subsequent iterations by sorting by weight, so as to better avoid missed detection and obtain more accurate multi-target detection results.

[0095] In the case where the number of targets decreases, the target particles with the highest probability of disappearance are filtered out through the second threshold, thereby better avoiding false detection and obtaining more accurate multi-target detection results.

[0096] According to any of the above embodiments, in this embodiment, the following is also included:

[0097] Determine a single target detection result of each target in the multi-target detection results according to the label;

[0098] Connecting the single target detection results of the same target at adjacent moments within the determination period to obtain the single target trajectory within the determination period;

[0099] Determining that the length of the single target trajectory is less than a third threshold, generating a false detection record;

[0100] If it is determined that there is a breakpoint in the single target trajectory, a missed detection record is generated.

[0101] According to the method of this embodiment, the motion trajectories of multiple targets can be further obtained, and false detections and missed detections can be verified based on the motion trajectories, thereby providing a more accurate multi-target recognition result.

[0102] Further, in a preferred embodiment, the step of predicting the response state set at the second moment according to the multi-target state set at the first moment comprises:

[0103] According to the multi-target state set at the first moment, predict the particle state based on the uniform linear motion model to obtain an intermediate prediction result;

[0104] According to the a priori prediction probability error range, an error disturbance term is added to the intermediate prediction result to obtain a response state set at the second moment.

[0105] The step of updating the weights of particles in the multi-target prediction state set at the second moment according to the measurement set detected by the vehicle-mounted millimeter-wave radar at the second moment comprises:

[0106] Normalizing the measurement set obtained by the detection of the vehicle-mounted millimeter-wave radar at the second moment to obtain a measurement normalized set;

[0107] According to the measurement normalization set, the weights of the particles in the multi-objective prediction state set at the second moment are updated.

[0108] In this embodiment, a more specific prediction and measurement method is provided, which can provide more accurate multi-target detection results while taking into account the prediction error.

[0109] According to any of the above embodiments, a detailed embodiment including a feasible prediction formula will be provided below in combination with the application scenario of an autonomous driving truck.

[0110] The self-driving truck is equipped with one forward-looking millimeter-wave radar sensor, two millimeter-wave corner radar sensors, and two rear-looking millimeter-wave radar sensors. In the millimeter-wave radar perception process, a large number of false detections are generated due to clutter and interference such as image interference and multipath interference. This poses a great challenge to the detection and tracking accuracy of the millimeter-wave radar perception algorithm. In the multi-target scenario, the multi-target detection and tracking algorithm based on millimeter-wave radar detection and tracking technology and random finite sets models the multiple measurements obtained by the multiple targets and radars separately, and distinguishes the states of different targets at the same time by introducing a label probability driving density (PHD) filter to achieve the purpose of multi-target detection and tracking.

[0111] The method of labeling Gaussian components in PHD (Probability Hypothesis Density) requires that the sensor likelihood function and Markov transition density are both linear Gaussian and the detection probability remains constant. During initialization, the particle set fitting the multi-target state space is divided, and particles with different labels are assigned to different targets, and the same target is assigned the same label. During the filtering process, the labels are predicted and updated simultaneously to distinguish different targets and give the motion trajectory of the target obstacle.

[0112] In particle PHD filtering, the particle set is predicted and updated as a specific implementation form of a random set. Finally, the target is extracted and detected in the filtered set. Each particle is described by the particle state and its corresponding weight. Here, an additional label identity variable is added to each particle. Its value can be a natural number. The value represents the track id of the particle. If the label value is 0, it means that the label belongs to the particle with a tentative source of detection. Here, we call such particles containing states, weights and labels as labeled particles, and the set of labeled particles used to fit the intensity of multiple targets is. The label here is only used to distinguish the ownership of the particles. It does not affect the propagation of PHD during the prediction and update process.

[0113] Steps for label PHD filtering:

[0114] Step 1, initialization.

[0115] In one processing cycle, the radar sensor echo is processed by signal processing to obtain a measurement set: that is, a point cloud set. The multi-target observation set is obtained through preprocessing, motion compensation, clustering algorithm, and post-processing algorithm. The state set of the multi-target is Gaussian sampling is performed on each target state in the set, and the particles obtained by sampling the same target are assigned the same label, and different targets are assigned different labels. At the same time, according to the initial weight distribution, the initialization label particle set is obtained.

[0116] Among them, N0 is the number of targets initialized (i.e. at the zeroth moment); is the state of the j-th target at the zeroth time (wherein the state includes any one or any combination of target speed, position, size, and Doppler value); are the label, weight, and state of the initialized (i.e., zeroth time) ith particle after Gaussian sampling; L0 is the number of initialized (i.e., zeroth time) particles, and the value of L0 is equal to the sum of the number of Gaussian samplings for N0 targets (for example, N0=3, 2, 3, and 3 particles are sampled for the 1st, 2nd, and 3rd targets respectively, then L0=8);

[0117] Since the particles obtained by sampling the same target in step 1 are assigned the same label, and the particles obtained by sampling different targets are assigned different labels, there are N0 values ​​in the label set. In this embodiment, for the particle obtained by sampling the jth target, its label value is set to j, so that 1≤τ j ≤N0.

[0118] Step 2: Prediction

[0119] Assume that the multi-target state PHD (probability hypothesis density) at time k-1 is composed of a weighted particle set To approximate, during the prediction process of the surviving particles, the labels corresponding to the particles remain unchanged, and the state prediction formula for the i-th target at the k-th moment refers to the following formula:

[0120]

[0121] q is the importance density function, p is the new chestnut importance density function

[0122]

[0123] In the formula, r represents the strength of the new random set; J is the number of new ones, and φ represents the state transition density

[0124] The corresponding particle set is New particles are sampled from the place where measurements were made at the previous moment, and a uniform provisional label τ = 0 is assigned to the new particles. Then the set of new box particles is The state and weight of the new particles also refer to the above two formulas. At this time, the total predicted particle set is:

[0125]

[0126] The predicted particle label set is

[0127] Step 3, Update

[0128] The measurements in the target measurement set obtained by the initial processing of the millimeter wave radar detection algorithm The calculated measurement normalization coefficient is

[0129]

[0130] P is the state prior probability, g is the likelihood function corresponding to the observation model, and W is the weight

[0131] For i=1,...,N+J k The predicted particle set, its state remains unchanged, the label value remains unchanged, and the weight is updated to

[0132]

[0133] P is the state prior, g is the likelihood, k is the noise intensity, and c is the observation statistical probability.

[0134] During the update process, the particle label remains unchanged, that is:

[0135]

[0136] Step 4: Target number estimation

[0137] Calculate the total number of targets Calculate the number of particles with the same label for each target Total target number estimate It can be obtained by summing the weights of the particle set that fits the multi-objective posterior PHD. k The weights of the box particles with the same label are summed as the number of single targets, so that the number of each target can be estimated. In the case of missed detection, it can be distinguished which target has not been detected.

[0138] Step 5: Label processing

[0139] When the target number estimate at time k is greater than the target number estimate at the previous time, that is, When , it means that the particles in the newborn box have detected the newborn of the target, and the number of newborns is Search for the box particles with label 0 whose weight is greater than the threshold χ and are in the top N from largest to smallest. new The labels of the particles in each box are changed to (τnewmax max , where τ max is the maximum label value of the box particles that have appeared.

[0140] When the target number estimate at time k is equal to the number estimate at the previous time, that is , it means there is no new target, and the label is not processed.

[0141] When the target number estimate at time k is less than the target number estimate at the previous time, that is, When , it indicates that there is no new target and there is a target disappearance. At this time, the weights and values ​​of each target calculated in step (4) are compared with the threshold χ. If the number of a target is less than the threshold, it means that the target disappears, and the labels of the corresponding box particles are all changed to 0. During the label processing, the weights and states of the box particles do not change. The processed label set is T k .

[0142] Step 6: Resampling

[0143] Particle set after label processing The particles with label 0 may be composed of two parts, one part is the newly detected new particles, and the other part is the particles corresponding to the disappeared targets. Neither of these two types is resampled. The resampling steps in Section 2 are performed on the set of particles with label value j and j≠0, and the resampled particle set corresponding to the target with label value j is obtained as in is the estimate of the number of targets with label value j. Similarly, we can get The resampled particles of the target, the resampled particle set is The number of particles is

[0144] Step 7: Target state estimation

[0145] According to the theory that particles with the same label originate from the same target, the particle set with label value j is directly clustered to obtain the state estimation of the target with label j. Similarly, all values ​​of the label in the resampled label set are traversed to obtain the state estimation of multiple targets, and the label of each target state remains unchanged.

[0146] Repeat steps 2 to 7 until the iteration ends.

[0147] Step 8: Target trajectory association and management

[0148] From the beginning of the observation time to the end of the observation time, the state estimates of the targets with the same label value are connected to form a string of data, which is the trajectory of the target. In turn, the trajectories of all targets that appear during the observation time can be obtained. If a track only appears for one moment and then disappears, it can be considered that the detection is a false detection; if a track is interrupted for one or two moments and then reappears, it can be considered that the two trajectories are caused by the missed detection of the same target.

[0149] The beneficial effects of this embodiment are:

[0150] By assigning labels to particles, target association can be avoided. At the same time, a continuous ID for each trajectory can be obtained without increasing the amount of calculation. Target tracking and trajectory maintenance can be achieved without increasing the complexity of the filter.

[0151] Compared with the method of representing the trajectory through an additional state variable, it has the advantages of reducing the complexity of the filter and saving computing resources.

[0152] The tag-based vehicle-mounted millimeter-wave radar multi-target detection device provided by the present invention is described below. The tag-based vehicle-mounted millimeter-wave radar multi-target detection device described below and the tag-based vehicle-mounted millimeter-wave radar multi-target detection method described above can be referenced to each other.

[0153] The embodiment of the present invention provides a label-based vehicle-mounted millimeter-wave radar multi-target detection system, including:

[0154] An acquisition module 1 is used to acquire a multi-target state set at a first moment;

[0155] Prediction module 2, used for predicting a response state set at a second moment according to the multi-objective state set at the first moment, and taking a union of the response state set at the second moment and the new state set as the multi-objective predicted state set;

[0156] The multi-target state set includes target state particles; the response state set includes response particles that survive or are derived from the target state particles; the new state set includes new particles that are sampled in the measurement area;

[0157] The target state particle includes a target state matrix, a label and a weight; the response particle includes a response state matrix, a label and a weight; the newborn particle includes a set newborn state matrix, a newborn label and a newborn weight;

[0158] A weight module 3, used to update the weights of particles in the multi-target prediction state set at the second moment according to the measurement set detected by the vehicle-mounted millimeter-wave radar at the second moment;

[0159] A label module 4, used for adjusting the labels of the particles in the multi-objective prediction state set at the second moment according to the updated weights of the particles in the multi-objective prediction state set at the second moment;

[0160] The detection module 5 is used to cluster the particles in the multi-target prediction state set at the second moment according to the labels, and perform weighted averaging on the clustering results according to the weights to obtain the multi-target detection result at the second moment.

[0161] Furthermore, the label module 4 includes:

[0162] An estimated value judgment unit is used to obtain an estimated value of the number of targets at the second moment according to the sum of the weights of the particles in the updated multi-target prediction state set at the second moment, and to judge the estimated value of the number of targets at the second moment:

[0163] Determine the difference N between the target quantity estimate at the second moment and the target quantity estimate at the first moment new is greater than zero, the label of the new target particle is adjusted to a value different from the label of the existing particle; the new target particle is a newly-born particle in the multi-target prediction state set at the second moment whose updated weight is greater than the first threshold, and the number of the new target particles is not greater than N new ;

[0164] Determine the difference N between the target quantity estimate at the second moment and the target quantity estimate at the first moment new If it is equal to zero, the existing label is kept;

[0165] Determine the difference N between the target quantity estimate at the second moment and the target quantity estimate at the first moment new is less than zero, the label of the disappeared target particle is adjusted to the same value as the new label; the disappeared target particle is a particle with the same label whose updated weight is less than the second threshold in the multi-target prediction state set at the second moment.

[0166] The acquisition module 1 comprises:

[0167] An exclusion unit, used to exclude particles with labels of new labels in the multi-target prediction state set at the first moment, to obtain a set to be resampled at the first moment;

[0168] A resampling unit is used to resample the multi-target state set at the first moment based on the set to be resampled at the first moment; in the multi-target state set at the first moment, the particle density in the set area is positively correlated with the particle weight in the set area in the set to be resampled at the first moment.

[0169] The prediction module 2 comprises:

[0170] An intermediate prediction unit, configured to predict the particle state based on the uniform linear motion model according to the multi-target state set at the first moment, to obtain an intermediate prediction result;

[0171] The disturbance unit is used to add an error disturbance term to the intermediate prediction result according to a priori prediction probability error range to obtain a response state set at a second moment.

[0172] The weight module 3 comprises:

[0173] A normalization unit, used to normalize a measurement set obtained by detection of the vehicle-mounted millimeter-wave radar at a second moment to obtain a measurement normalization set;

[0174] A weight prediction unit is used to update the weights of particles in the multi-target prediction state set at the second moment according to the measurement normalization set.

[0175] Furthermore, the system further comprises:

[0176] A single target module, used to determine a single target detection result of each target in the multi-target detection results according to the label;

[0177] A trajectory module, used to connect the single target detection results of the same target at adjacent moments within a determination period to obtain the single target trajectory within the determination period;

[0178] a false detection module, configured to generate a false detection record if the length of the single target trajectory is less than a third threshold;

[0179] The missed detection module is used to determine if there is a breakpoint in the single target trajectory and generate a missed detection record.

[0180] Figure 3 An example of a physical structure diagram of an electronic device is shown in FIG. Figure 3As shown, the electronic device may include: a processor 310, a communication interface 320, a memory 330 and a communication bus 340, wherein the processor 310, the communication interface 320 and the memory 330 communicate with each other through the communication bus 340. The processor 310 may call the logic instructions in the memory 330 to execute the tag-based vehicle-mounted millimeter-wave radar multi-target detection method, the method comprising: obtaining a multi-target state set at a first moment; predicting a response state set at a second moment according to the multi-target state set at the first moment, and taking the union of the response state set at the second moment and the new state set as the multi-target predicted state set; the multi-target state set includes target state particles; the response state set includes response particles obtained by surviving or deriving the target state particles; the new state set includes new particles sampled in the measurement area; the target state particles include a target state matrix , labels and weights; the response particles include a response state matrix, labels and weights; the new particles include a set new state matrix, new labels and new weights; according to the measurement set obtained by the detection of the on-board millimeter-wave radar at the second moment, the weights of the particles in the multi-target prediction state set at the second moment are updated; according to the updated weights of the particles in the multi-target prediction state set at the second moment, the labels of the particles in the multi-target prediction state set at the second moment are adjusted; the particles in the multi-target prediction state set at the second moment are clustered according to the labels, and the clustering results are weighted averaged according to the weights to obtain the multi-target detection result at the second moment.

[0181] In addition, the logic instructions in the above-mentioned memory 330 can be implemented in the form of a software functional unit and can be stored in a computer-readable storage medium when it is sold or used as an independent product. Based on such an understanding, the technical solution of the present invention, in essence, or the part that contributes to the prior art or the part of the technical solution, can be embodied in the form of a software product, and the computer software product is stored in a storage medium, including a number of instructions for a computer device (which can be a personal computer, a server, or a network device, etc.) to perform all or part of the steps of the method described in each embodiment of the present invention. The aforementioned storage medium includes: U disk, mobile hard disk, read-only memory (ROM, Read-Only Memory), random access memory (RAM, Random Access Memory), disk or optical disk and other media that can store program codes.

[0182] On the other hand, the present invention also provides a computer program product, which includes a computer program. The computer program can be stored on a non-transitory computer-readable storage medium. When the computer program is executed by a processor, the computer can execute the label-based vehicle-mounted millimeter-wave radar multi-target detection method provided by the above-mentioned methods, and the method includes: obtaining a multi-target state set at a first moment; predicting a response state set at a second moment based on the multi-target state set at the first moment, and taking the union of the response state set at the second moment and the new state set as the multi-target predicted state set; the multi-target state set includes target state particles; the response state set includes response particles that survive or are derived from the target state particles; the new state set includes The method comprises the following steps: enclosing new particles sampled within the measurement area; the target state particles include a target state matrix, a label and a weight; the response particles include a response state matrix, a label and a weight; the new particles include a set new state matrix, a new label and a new weight; according to the measurement set obtained by the detection of the on-board millimeter-wave radar at the second moment, updating the weights of the particles in the multi-target prediction state set at the second moment; adjusting the labels of the particles in the multi-target prediction state set at the second moment according to the updated weights of the particles in the multi-target prediction state set at the second moment; clustering the particles in the multi-target prediction state set at the second moment according to the labels, and performing weighted averaging on the clustering results according to the weights to obtain the multi-target detection result at the second moment.

[0183] On the other hand, the present invention also provides a non-transitory computer-readable storage medium having a computer program stored thereon, which is implemented when the processor executes the label-based vehicle-mounted millimeter-wave radar multi-target detection method provided by the above-mentioned methods, the method comprising: obtaining a multi-target state set at a first moment; predicting a response state set at a second moment based on the multi-target state set at the first moment, and taking the union of the response state set at the second moment and the new state set as the multi-target predicted state set; the multi-target state set includes target state particles; the response state set includes response particles that survive or are derived from the target state particles; the new state set includes new particles sampled in the measurement area Particles; the target state particles include a target state matrix, a label and a weight; the response particles include a response state matrix, a label and a weight; the new particles include a set new state matrix, a new label and a new weight; according to the measurement set obtained by the detection of the on-board millimeter-wave radar at the second moment, the weights of the particles in the multi-target prediction state set at the second moment are updated; according to the updated weights of the particles in the multi-target prediction state set at the second moment, the labels of the particles in the multi-target prediction state set at the second moment are adjusted; the particles in the multi-target prediction state set at the second moment are clustered according to the labels, and the clustering results are weighted averaged according to the weights to obtain the multi-target detection result at the second moment.

[0184] The device embodiments described above are merely illustrative, wherein the units described as separate components may or may not be physically separated, and the components displayed as units may or may not be physical units, that is, they may be located in one place, or they may be distributed on multiple network units. Some or all of the modules may be selected according to actual needs to achieve the purpose of the scheme of this embodiment. Ordinary technicians in this field can understand and implement it without paying creative labor.

[0185] Through the description of the above implementation methods, those skilled in the art can clearly understand that each implementation method can be implemented by means of software plus a necessary general hardware platform, and of course, it can also be implemented by hardware. Based on this understanding, the above technical solution is essentially or the part that contributes to the prior art can be embodied in the form of a software product, and the computer software product can be stored in a computer-readable storage medium, such as ROM / RAM, a disk, an optical disk, etc., including a number of instructions for a computer device (which can be a personal computer, a server, or a network device, etc.) to execute the methods described in each embodiment or some parts of the embodiments.

[0186] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, rather than to limit it. Although the present invention has been described in detail with reference to the aforementioned embodiments, those skilled in the art should understand that they can still modify the technical solutions described in the aforementioned embodiments, or make equivalent replacements for some of the technical features therein. However, these modifications or replacements do not deviate the essence of the corresponding technical solutions from the spirit and scope of the technical solutions of the embodiments of the present invention.

Claims

1. A label-based vehicle-mounted millimeter-wave radar multi-target detection method, characterized in that: include: Get the multi-target state set at the first moment; Predicting a response state set at a second moment according to the multi-objective state set at the first moment, and taking the union of the response state set at the second moment and the new state set as the multi-objective predicted state set; The multi-target state set includes target state particles; the response state set includes response particles that survive or are derived from the target state particles; the new state set includes new particles that are sampled in the measurement area; The target state particle includes a target state matrix, a label and a weight; the response particle includes a response state matrix, a label and a weight; the newborn particle includes a set newborn state matrix, a newborn label and a newborn weight; According to the measurement set obtained by the detection of the vehicle-mounted millimeter-wave radar at the second moment, updating the weights of the particles in the multi-target prediction state set at the second moment; Adjusting the labels of the particles in the multi-objective prediction state set at the second moment according to the updated weights of the particles in the multi-objective prediction state set at the second moment; The particles in the multi-target prediction state set at the second moment are clustered according to the labels, and the clustering results are weighted averaged according to the weights to obtain the multi-target detection result at the second moment.

2. The label-based vehicle-mounted millimeter-wave radar multi-target detection method according to claim 1 is characterized in that: The step of adjusting the labels of the particles in the multi-objective prediction state set at the second moment according to the updated weights of the particles in the multi-objective prediction state set at the second moment comprises: According to the sum of the weights of the particles in the updated multi-objective prediction state set at the second moment, the estimated value of the number of targets at the second moment is obtained, and the estimated value of the number of targets at the second moment is judged: Determine the difference N between the target quantity estimate at the second moment and the target quantity estimate at the first moment new is greater than zero, the label of the new target particle is adjusted to a value different from the label of the existing particle; the new target particle is a newly-born particle in the multi-target prediction state set at the second moment whose updated weight is greater than the first threshold, and the number of the new target particles is not greater than N new ; Determine the difference N between the target quantity estimate at the second moment and the target quantity estimate at the first moment new If it is equal to zero, the existing label is kept; Determine the difference N between the target quantity estimate at the second moment and the target quantity estimate at the first moment new is less than zero, the label of the disappeared target particle is adjusted to the same value as the new label; the disappeared target particle is a particle with the same label whose updated weight is less than the second threshold in the multi-target prediction state set at the second moment.

3. The label-based vehicle-mounted millimeter-wave radar multi-target detection method according to claim 2 is characterized in that: The step of obtaining a multi-target state set at a first moment comprises: Excluding particles with new labels in the multi-target prediction state set at the first moment, and obtaining the set to be resampled at the first moment; Based on the set to be resampled at the first moment, the multi-target state set at the first moment is resampled; in the multi-target state set at the first moment, the particle density in the set area is positively correlated with the particle weight in the set area in the set to be resampled at the first moment.

4. The tag-based vehicle-mounted millimeter-wave radar multi-target detection method according to claim 1, characterized in that: Also includes: Determine a single target detection result of each target in the multi-target detection results according to the label; Connecting the single target detection results of the same target at adjacent moments within the determination period to obtain the single target trajectory within the determination period; Determining that the length of the single target trajectory is less than a third threshold, generating a false detection record; If it is determined that there is a breakpoint in the single target trajectory, a missed detection record is generated.

5. The tag-based vehicle-mounted millimeter-wave radar multi-target detection method according to claim 1, characterized in that: The step of predicting a response state set at a second moment according to the multi-target state set at the first moment comprises: According to the multi-target state set at the first moment, predict the particle state based on the uniform linear motion model to obtain an intermediate prediction result; According to the a priori prediction probability error range, an error disturbance term is added to the intermediate prediction result to obtain a response state set at the second moment.

6. The tag-based vehicle-mounted millimeter-wave radar multi-target detection method according to claim 1, characterized in that: The step of updating the weights of particles in the multi-target prediction state set at the second moment according to the measurement set detected by the vehicle-mounted millimeter-wave radar at the second moment comprises: Normalizing the measurement set obtained by the detection of the vehicle-mounted millimeter-wave radar at the second moment to obtain a measurement normalized set; According to the measurement normalization set, the weights of the particles in the multi-objective prediction state set at the second moment are updated.

7. A label-based vehicle-mounted millimeter-wave radar multi-target detection system, characterized in that: include: An acquisition module, used to acquire a multi-target state set at a first moment; A prediction module, configured to predict a response state set at a second moment according to the multi-objective state set at the first moment, and use a union of the response state set at the second moment and the new state set as the multi-objective prediction state set; The multi-target state set includes target state particles; the response state set includes response particles that survive or are derived from the target state particles; the new state set includes new particles that are sampled in the measurement area; The target state particle includes a target state matrix, a label and a weight; the response particle includes a response state matrix, a label and a weight; the newborn particle includes a set newborn state matrix, a newborn label and a newborn weight; A weight module, used to update the weights of particles in the multi-target prediction state set at the second moment according to the measurement set detected by the vehicle-mounted millimeter-wave radar at the second moment; A label module, used for adjusting the labels of the particles in the multi-objective prediction state set at the second moment according to the updated weights of the particles in the multi-objective prediction state set at the second moment; The detection module is used to cluster the particles in the multi-target prediction state set at the second moment according to the labels, and perform weighted averaging on the clustering results according to the weights to obtain the multi-target detection result at the second moment.

8. An electronic device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that: When the processor executes the program, the steps of the label-based vehicle-mounted millimeter-wave radar multi-target detection method as described in any one of claims 1 to 6 are implemented.

9. A non-transitory computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the steps of the label-based vehicle-mounted millimeter-wave radar multi-target detection method as described in any one of claims 1 to 6 are implemented.

10. A computer program product, comprising a computer program, characterized in that When the computer program is executed by a processor, the steps of the label-based vehicle-mounted millimeter-wave radar multi-target detection method as described in any one of claims 1 to 6 are implemented.

Images