Multi-sensor probability hypothesis density multi-target tracking method

Abstract

The invention discloses a multi-target tracking method for multi-sensor probability hypothesis density. The method comprises the following steps: establishing a tracking model for the multi-sensor probability hypothesis density; performing state recursion on the probability hypothesis density of the multiple sensors according to the tracking model; solving the prediction intensity of the multi-sensor probability hypothesis density; solving the posterior intensity of the probability hypothesis density of the multiple sensors; and pruning and merging the probability hypothesis density of the multiple sensors, and estimating a target state and deviation. Probability hypothesis density recursion is applied by introducing a motion model of translation measurement deviation into a related intensity function. Under a linear Gaussian hypothesis on a deviation motion model, Gaussian mixture implementation is used to give an expression in a closed form. The target state and the translation measurement deviation are coupled through likelihood in the updating step, so that a two-stage Kalman filter is adopted to approximate an easy-to-process form, and the calculation complexity is greatly reduced.

Description

Technical Field

[0001] The invention relates to the technical field of information fusion, and in particular to a multi-target tracking method of multi-sensor probability hypothesis density. Background Art

[0002] Registration error compensation has always been an important issue in multi-sensor data fusion systems, regardless of whether the sensor measurements are processed in a centralized or distributed manner. There are many kinds of sensor biases, such as translation bias in state space, rotation bias in state space, translation bias in measurement space, rotation bias in measurement space, and translation and rotation bias. The estimation of unknown translational measurement biases has received great attention, and it is this problem that we address in this paper. Note that it is crucial to estimate these measurement biases as accurately as possible so that the multi-sensor measurements can be referenced to a common tracking coordinate frame. To address this issue, many methods have been proposed, such as least squares, maximum likelihood, and Kalman filtering, among which Kalman filtering has always been an attractive method for its efficiency.

[0003] Recently, the theory of finite set statistics (FISST) has been used to solve the problem of multi-target tracking with data association avoided. In the framework of FISST, the target states and measurements are modeled as two different random finite sets (RFSs). Therefore, the problem of tracking an unknown and time-varying number of targets in a cluttered environment can be solved. In addition, by constructing multi-target transition density and multi-target likelihood functions, multi-target tracking can be formulated in a strict Bayesian framework. However, due to the combinatorial nature of multi-set integrals and multi-target densities, the optimal multi-target Bayesian filter is usually intractable. To alleviate this intractability, probability hypothesis density (PHD) filters have been proposed as first-order moment approximations of the multi-target posterior density. It should be pointed out that the PHD recursion still requires solving multi-dimensional integrals. Summary of the invention

[0004] In view of the above problems, the present invention is proposed to provide a multi-target tracking method of multi-sensor probability hypothesis density that overcomes the above problems or at least partially solves the above problems.

[0005] In order to solve the above technical problems, the embodiments of the present application disclose the following technical solutions:

[0006] In a first aspect, an embodiment of the present invention discloses a multi-target tracking method of multi-sensor probability hypothesis density, comprising:

[0007] S100. Establishing a tracking model for multi-sensor probability hypothesis density;

[0008] S200. Perform state recursion on the multi-sensor probability hypothesis density according to the tracking model;

[0009] S300. Obtaining the predicted strength of the multi-sensor probability hypothesis density;

[0010] S400. Obtaining the posterior strength of the multi-sensor probability hypothesis density;

[0011] S500. Prune and merge the probability hypothesis density of multiple sensors to estimate the target state and deviation.

[0012] Furthermore, in S100, a tracking model is established for the multi-sensor probability hypothesis density, wherein the tracking model includes a linear target motion model and a measurement model, wherein the linear target motion model and the measurement model are respectively set as: x k =F k- 1x k-1 +w k-1 , l=1,2,...,L, where and denote the target state and the lth sensor measurement respectively. L is the number of sensors. k-1 and is the state transfer matrix and the measurement matrix. k-1 and They are the covariance matrices and Zero-mean white Gaussian process noise and measurement noise. is the translation measurement deviation of the lth sensor.

[0013] Furthermore, in S200, the multi-sensor probability hypothesis density is state recursively performed according to the tracking model, and the specific method includes: using PHD recursion as the optimal multi-objective Bayesian recursion, by enhancing the state vector in Remember v k-1k-1 (x k-1 ,b k-1 ) is the intensity posterior function at time k-1, then the predicted intensity function v is derived kk-1 (x k ,b k ), assuming that the registration errors of different sensors are independent, that is, Said is the transfer strength of the lth sensor registration error, and the registration error It can be described as a first-order Gaussian Markov process, that is, After all sensor measurement data are obtained, the posterior strength v at time k can be obtained by sequential processing. kk .

[0014] Furthermore, the calculation formula of the prediction strength function is:

[0015] v kk-1 (x k ,b k )=∫[p s f x (x k |x k-1 )f(b k |b k-1 )+β kk-1 (x k ,b k |x k-1 ,b k-1 )]·v k-1k-1 (x k-1 ,b k-1 )dx k-1 db k-1 +γ k (x k ,b k )

[0016] Where p s is the survival probability of the existing target, f x and f b are the single target state transfer function and the registration error transfer density respectively; β kk-1 and γ k They represent the intensity function of the derived target random finite set and the intensity function of the outside world entering the new target random finite set respectively.

[0017] Further, in S300, the prediction strength of the multi-sensor probability hypothesis density is obtained, and the specific method includes: assuming that the random finite set PHD and the derived random finite set PHD of the new target entering the outside world are expressed in the form of a Gaussian mixture, that is:

[0018]

[0019] In the formula, J γ,k , as well as They are respectively the number of freshmen, freshmen weight, freshmen mean, freshmen deviation, deviation covariance, and freshmen covariance that determine the random finite set PHD of the new target from the outside world;

[0020] J β,k , as well as To determine the number of derivatives, derivative weights, derivative transfer functions, derivative noise, derivative deviation transfer functions, and derivative deviation covariance of the random finite set PHD of the derived new target;

[0021] Assume that the posterior strength function v at time k-1 is k-1k-1 (x,b) is in the form of a Gaussian mixture, that is:

[0022]

[0023] Where: J is the number of Gaussian terms; w is the weight of the jth Gaussian term, then the predicted intensity function v kk-1 (x k ,b k )for

[0024] v kk-1 (x k ,b k )=v s,kk-1 (x k ,b k )+v β,kk-1 (x k ,b k )+v γ,kk-1 (x k ,b k )

[0025] In the formula, v s,kk-1 (x k ,b k ) is the survival strength, v β,kk-1 (x k ,b k ) is the derived strength, v γ,kk-1 (x k ,b k ) is the new strength.

[0026] Furthermore, the posterior strength function at time k is specifically:

[0027]

[0028] In the formula,

[0029] in, is the detection probability of the lth sensor; h l (z k |x x ,b k ) is the single target measurement likelihood function of the lth sensor; is the intensity function of a random finite set of clutter.

[0030] Furthermore, in S400, the posterior strength of the multi-sensor probability hypothesis density is obtained, and the specific method includes:

[0031] The prediction strength function is recorded as:

[0032]

[0033] Then the update intensity function after multi-sensor sequential processing is:

[0034]

[0035] Furthermore, the Kalman prediction process at time k is as follows:

[0036]

[0037] Furthermore, the Kalman update process at time k is as follows:

[0038]

[0039] In the formula is the likelihood function of sensor l, is the new information of sensor l, is the predicted position of sensor l, is the measurement covariance of sensor l, is the measurement transfer function of sensor l.

[0040] In a second aspect, an embodiment of the present invention discloses an electronic device, including:

[0041] one or more processors;

[0042] A memory for storing one or more programs;

[0043] When the one or more programs are executed by the one or more processors, the one or more processors implement the multi-target tracking method.

[0044] The beneficial effects of the above technical solution provided by the embodiment of the present invention include at least:

[0045] The present invention discloses a multi-target tracking method of multi-sensor probability hypothesis density, comprising: establishing a tracking model for the multi-sensor probability hypothesis density; performing state recursion on the multi-sensor probability hypothesis density according to the tracking model; obtaining the predicted strength of the multi-sensor probability hypothesis density; obtaining the posterior strength of the multi-sensor probability hypothesis density; pruning and merging the multi-sensor probability hypothesis density to estimate the target state and deviation. The present invention solves the multi-sensor multi-target tracking problem with registration errors in random finite set formulation. The probability hypothesis density probability hypothesis density (PHD) recursion is applied by introducing the motion model of the translation measurement deviation into the relevant intensity function. Under the linear Gaussian assumption on the deviation motion model, a Gaussian mixture implementation is used to give a closed form expression. Since the target state and the translation measurement deviation are coupled through the likelihood in the update step, a two-stage Kalman filter is used to approximate a tractable form, thereby greatly reducing the computational complexity. An example simulation is provided to verify the effectiveness of the proposed filter.

[0046] The technical solution of the present invention is further described in detail below through the accompanying drawings and embodiments. BRIEF DESCRIPTION OF THE DRAWINGS

[0047] The accompanying drawings are used to provide a further understanding of the present invention and constitute a part of the specification. Together with the embodiments of the present invention, they are used to explain the present invention and do not constitute a limitation of the present invention. In the accompanying drawings:

[0048] Figure 1 This is a flowchart of a multi-target tracking method of multi-sensor probability hypothesis density in embodiment 1 of the present invention;

[0049] Figure 2 is a schematic diagram of motion trajectories and estimation of multiple maneuvering targets in Embodiment 2 of the present invention;

[0050] Figure 3 is a schematic diagram of motion trajectories and estimation of multiple maneuvering targets in Embodiment 2 of the present invention;

[0051] Figure 4 This is a schematic diagram of OSPA distances of two algorithms in Example 3 of the present invention. DETAILED DESCRIPTION

[0052] The exemplary embodiments of the present disclosure will be described in more detail below with reference to the accompanying drawings. Although the exemplary embodiments of the present disclosure are shown in the accompanying drawings, it should be understood that the present disclosure can be implemented in various forms and should not be limited by the embodiments set forth herein. On the contrary, these embodiments are provided to enable a more thorough understanding of the present disclosure and to fully convey the scope of the present disclosure to those skilled in the art.

[0053] In order to solve the problems existing in the prior art, an embodiment of the present invention provides a multi-target tracking method of multi-sensor probability hypothesis density.

[0054] Example 1

[0055] The present invention discloses a multi-target tracking method with multi-sensor probability hypothesis density. Figure 1 ,include:

[0056] S100. Establish a tracking model for multi-sensor probability hypothesis density; In complex environments, a single sensor often cannot provide sufficiently accurate and stable information for target tracking. Therefore, multi-sensor information fusion becomes the key to improving tracking performance. Specifically, in the multi-sensor fusion tracking problem, probability hypothesis density (PHD) is a commonly used state recursion method, which can estimate the target in a dynamic environment and can handle multiple targets and observation data from different sensors. The PHD filter updates the target state by recursively recursively, and its main advantage is that it can handle data association between sensors and multi-target situations.

[0057] In S100 of this embodiment, a tracking model is established for the multi-sensor probability hypothesis density, and the tracking model includes a linear target motion model and a measurement model, wherein the linear target motion model and the measurement model are respectively set as: x k =F k-1 x k-1 +w k-1 , l=1,2,...,L, where and denote the target state and the lth sensor measurement respectively. L is the number of sensors. k-1 and is the state transfer matrix and the measurement matrix. k-1 and They are the covariance matrices and Zero-mean white Gaussian process noise and measurement noise. is the translation measurement deviation of the lth sensor.

[0058] Specifically, in a multi-sensor tracking system, the measurements received from each sensor should be converted to a common coordinate system for merging, and often encounter registration errors, which also refers to the estimation of unknown measurement deviations. Registration errors have many forms, such as translation errors and rotation errors in state space, translation errors and rotation errors in measurement space, etc. Consider the linear target motion model and the measurement model set as: k =F k-1 x k-1 +w k-1 , l=1,2,...,L, where and denote the target state and the lth sensor measurement respectively. L is the number of sensors. k-1 and is the state transfer matrix and the measurement matrix. k-1 and They are the covariance matrices and Zero-mean white Gaussian process noise and measurement noise. is the translation measurement deviation of the lth sensor.

[0059] S200. Perform state recursion on the multi-sensor probability hypothesis density according to the tracking model; In S200 of this embodiment, the state recursion on the multi-sensor probability hypothesis density according to the tracking model is performed, and the specific method includes: using PHD recursion as the optimal multi-objective Bayesian recursion, by enhancing the state vector in Remember v k-1k-1 (x k-1 ,b k-1 ) is the intensity posterior function at time k-1, then the predicted intensity function v is derived kk-1 (x k ,b k ), assuming that the registration errors of different sensors are independent, that is, Said is the transfer strength of the lth sensor registration error, and the registration error It can be described as a first-order Gaussian Markov process, that is, After all sensor measurement data are obtained, the posterior strength v at time k can be obtained by sequential processing. kk .

[0060] Specifically, by using the theory of finite set statistics, the optimal Bayesian recursion can be obtained based on the multi-target posterior density function. However, this recursion involves multiple integrals, and the multi-target posterior density function is combinatorial, which makes it computationally intractable. To alleviate this intractability, inspired by the single target tracking method, the propagation of statistical moments associated with the posterior density is adopted. By propagating the first-order moments or intensity functions of random finite sets of multi-targets, the PHD recursion provides a computationally cheaper alternative to the optimal multi-target Bayesian recursion. Due to the registration error b k l is unknown and needs to be estimated jointly with the target state, similar to the derivation of the standard PHD recursion, so we can enhance the state vector in Remember v k-1k-1 (x k-1 ,bk-1 ) is the intensity posterior function at time k-1, then the predicted intensity function v can be derived kk-1 (x k ,b k ):

[0061] v kk-1 (x k ,b k )=∫[p s f x (x k |x k-1 )f b (b k |b k-1 )+β kk-1 (x k ,b k |x k-1 ,b k-1 )]·v k-1k-1 (x k-1 ,b k-1 )dx k-1 db k-1 +γ k (x k ,b k )

[0062] Where p s is the survival probability of the existing target, f x and f b are the single target state transfer function and the registration error transfer density respectively; β kk-1 and γ k They represent the intensity function of the derived target random finite set and the intensity function of the outside world entering the new target random finite set respectively.

[0063] Assume that the registration errors of different sensors are independent, that is, In the above formula, is the transfer strength of the lth sensor registration error, and the registration error It can be described as a first-order Gaussian Markov process, that is, After all sensor measurement data are obtained, the posterior strength v at time k can be obtained by sequential processing. kk :

[0064]

[0065] In the formula,

[0066] in, is the detection probability of the lth sensor; h l (z k |x x ,bk ) is the single target measurement likelihood function of the lth sensor; is the intensity function of a random finite set of clutter.

[0067] S300. Obtain the prediction strength of the multi-sensor probability hypothesis density; in a multi-sensor probability hypothesis density (PHD) filter, the prediction strength (also called the prediction probability density or the expected value of the number of targets) refers to the probability distribution of the target at the next moment, given the observation information at the current moment and the previous state estimation. This process involves a dynamic model of the target state and the fusion of multi-sensor observations.

[0068] Specifically, in order to obtain the PHD recursive solution in the form of Gaussian mixture, it is assumed that the random finite set PHD and the derived random finite set PHD of the new target can be expressed in the form of Gaussian mixture, that is,

[0069]

[0070] In the formula, J γ,k , as well as They are the number of new students, new student weights, new student means, new student deviations, deviation covariance, and new student covariance that determine the random finite set PHD of the new target from the outside world; J β,k , as well as To determine the number of derivatives, derivative weights, derivative transfer functions, derivative noise, derivative deviation transfer functions, and derivative deviation covariance of the random finite set PHD of the derived new targets.

[0071] Based on the above assumptions, the following implementation form of the recursive formula can be obtained:

[0072] Assume that the posterior strength function v at time k-1 is k-1k-1 (x,b) is in the form of a Gaussian mixture, that is:

[0073]

[0074] Where: J is the number of Gaussian terms; w is the weight of the jth Gaussian term, then the predicted intensity function v kk-1 (x k ,b k )for

[0075] v kk-1 (x k ,b k )=v s,kk-1 (x k ,b k )+v β,kk-1 (x k ,b k )+vγ,kk-1 (x k ,b k )

[0076] In the formula, v s,kk-1 (x k ,b k ) is the survival strength, v β,kk-1 (x k ,b k ) is the derived strength, v γ,kk-1 (x k ,b k ) is the new strength.

[0077]

[0078] Through the above steps, the prediction strength of the target state in a multi-sensor environment can be obtained, which provides a basis for subsequent target tracking and updating.

[0079] S400. Obtain the posterior strength of the multi-sensor probability hypothesis density; in a multi-sensor probability hypothesis density (PHD) filter, the posterior strength is the updated probability density of the target state, which combines the sensor observation data with the predicted PHD to estimate the state of the target. The posterior strength reflects the expected density of the target in a specific state (such as position) after given the observation information.

[0080] In S400 of this embodiment, the posterior strength of the multi-sensor probability hypothesis density is obtained, and the specific method includes:

[0081] The prediction strength function is recorded as:

[0082]

[0083] Then the update intensity function after multi-sensor sequential processing is:

[0084]

[0085] Among them, the Kalman prediction process at time k is specifically:

[0086]

[0087] In the formula is the likelihood function of sensor l, is the new information of sensor l, is the predicted position of sensor l, is the measurement covariance of sensor l, is the measurement transfer function of sensor l.

[0088] S500. Prune and merge the probability hypothesis density of multiple sensors to estimate the target state and deviation. In the multi-sensor PHD filtering process, since each sensor generates a large amount of measurement data, the number of elements in the target state set increases sharply. In order to reduce the computational complexity and maintain the real-time tracking, it is necessary to prune the target state set. The pruning operation is usually performed based on the probability hypothesis density (or weight) of the target state. When the weight of a target state is lower than a preset threshold, it is considered that the target state contributes less to the tracking result and can be removed from the target state set. After the pruning operation, there may still be multiple similar target states. In order to further improve the accuracy and robustness of tracking, it is necessary to merge these similar target states. The merging operation is usually performed based on the distance or similarity between the target states. When the distance between two or more target states is less than a preset threshold, they are considered to represent the same target and can be merged into one target state.

[0089] The present embodiment discloses a multi-target tracking method of multi-sensor probability hypothesis density, including: establishing a tracking model for the multi-sensor probability hypothesis density; performing state recursion on the multi-sensor probability hypothesis density according to the tracking model; obtaining the predicted strength of the multi-sensor probability hypothesis density; obtaining the posterior strength of the multi-sensor probability hypothesis density; pruning and merging the multi-sensor probability hypothesis density to estimate the target state and deviation. The present embodiment solves the multi-sensor multi-target tracking problem with registration errors in the random finite set formulation. The probability hypothesis density probability hypothesis density (PHD) is recursively applied by introducing the motion model of the translation measurement deviation into the relevant intensity function. Under the linear Gaussian assumption on the deviation motion model, a Gaussian mixture implementation is used to give a closed form expression. Since the target state and the translation measurement deviation are coupled through the likelihood in the update step, a two-stage Kalman filter is used to approximate a tractable form, thereby greatly reducing the computational complexity. An example simulation is provided to verify the effectiveness of the proposed filter.

[0090] Example 2

[0091] In order to verify the method disclosed in Example 1, this example verifies this method. The algorithm and processing method of Example 1 have been verified and achieved satisfactory application results:

[0092] 1. Experimental conditions: Assume that the random finite set PHD of each sensor clutter still uses the Gaussian mixture model. In the simulation process, the motion state of the target can be expressed as {x,v x ,y,v y}, where {x, y} are the position information of the target in the X and Y directions in the horizontal rectangular coordinate system, respectively, and {vx ,v y} are the speed information of the target in the X and Y directions in the horizontal rectangular coordinate system, (s x ,s y ) represents the position of the sensor. The sampling time interval is T = 1s; the simulation time is 100s; the target survival probability is p S =0.99; the probability of being detected is taken as p D =0.98; To avoid the exponential growth of the number of Gaussian terms, a pruning and merging strategy is adopted, where the discard threshold is set to T m =10 -3 , the merging threshold is taken as U Th =5, the Gaussian component with a weight greater than 0.5 in the Gaussian mixture term of the intensity function is taken as the target state estimate, that is, ω Th = 0.5; to limit the number of Gaussian terms in each iteration, it is assumed that it does not exceed 100 terms, that is, J max =100,w=0.

[0093]

[0094] 2. Simulation content:

[0095] Considering a two-dimensional multi-target tracking scenario, a coordinated turning model is used to describe the target's motion. In this example, two sensors are used to track multiple maneuvering targets. The vectors are still generated by the distance and azimuth equations. Two sensors are used to track multiple maneuvering targets. The positions of the sensors are (2,15) (in km) and (18,15) (in km). The measurement noise V2 is modeled as zero-mean Gaussian white noise and the covariance matrix is The registration errors of the two sensors are (500,π / 360) and (400,π / 120) respectively.

[0096] Assume that new targets entering from the outside world may appear from three regions, whose random finite sets are Poisson, and whose PHD is

[0097]

[0098] Where:

[0099]

[0100] Assume that the random finite set of derived targets is Poisson and the PHD is

[0101]

[0102] Where: Q x,k =diag{10 4,400,10 4 ,400};

[0103] 3. Analysis of simulation results:

[0104] Figure 2 The real motion trajectories of the four targets in the simulation process are given. The target motion trajectories are as follows: Target 1 moves at a constant speed for 100 seconds at the initial position (10, 20) (unit: km) at the first moment; Target 2 is derived from Target 1 at the 30th second and moves at a constant speed for 40 seconds; Target 3 starts to move at the initial position (0, 30) (unit: km) at the 5th second and moves at a constant speed for 95 seconds; Target 4 is derived from Target 3 at the 40th second and moves at a constant speed for 40 seconds. Figure 2 As shown, multi-target tracking can be well achieved using the multi-sensor registration algorithm (GM-PHD-RE). Figure 3 As shown in the figure, using OSPA distance as the evaluation index, compared with the multi-target tracking without processing the registration error (GM-PHD), the proposed multi-target tracking algorithm with processing the registration error has better tracking performance.

[0105] The simulation results demonstrate the effectiveness of the multi-target tracking method based on multi-sensor probability hypothesis density proposed in this patent.

[0106] Example 3

[0107] Based on the same inventive concept, an embodiment of the present disclosure also provides an electronic device. Figure 4 FIG. 1 is a schematic diagram of the structure of an electronic device according to an embodiment of the present disclosure. Figure 4 As shown, an embodiment of the present disclosure provides an electronic device including: one or more processors 101, a memory 102, and one or more I / O interfaces 103. The memory 102 stores one or more programs, and when the one or more programs are executed by the one or more processors, the one or more processors implement any optimization method in the above embodiments; the one or more I / O interfaces 103 are connected between the processor and the memory, and are configured to implement information interaction between the processor and the memory.

[0108] Among them, the processor 101 is a device with data processing capabilities, including but not limited to a central processing unit (CPU), etc.; the memory 102 is a device with data storage capabilities, including but not limited to random access memory (RAM, more specifically SDRAM, DDR, etc.), read-only memory (ROM), electrically erasable programmable read-only memory (EEPROM), flash memory (FLASH); the I / O interface (read-write interface) 103 is connected between the processor 101 and the memory 102, and can realize information interaction between the processor 101 and the memory 102, including but not limited to a data bus (Bus), etc.

[0109] In some embodiments, the processor 101 , the memory 102 , and the I / O interface 103 are connected to each other via a bus 104 , and further connected to other components of the computing device.

[0110] In some embodiments, the one or more processors 101 include a field programmable gate array.

[0111] According to an embodiment of the present disclosure, a computer-readable medium is further provided, wherein a computer program is stored on the computer-readable medium, wherein when the program is executed by a processor, the steps in any optimization method in the above-mentioned embodiment are implemented.

[0112] It should be understood that the specific order or hierarchy of steps in the disclosed process is an example of an exemplary method. Based on design preferences, it should be understood that the specific order or hierarchy of steps in the process can be rearranged without departing from the scope of protection of the present disclosure. The attached method claims present the elements of the various steps in an exemplary order and are not intended to be limited to the specific order or hierarchy described.

[0113] In the above detailed description, various features are grouped together in a single embodiment to simplify the disclosure. This method of disclosure should not be interpreted as reflecting an intention that the embodiments of the claimed subject matter require more features than are clearly stated in each claim. On the contrary, as reflected in the appended claims, the invention is in a state of having less than all the features of the disclosed individual embodiments. Therefore, the appended claims are hereby expressly incorporated into the detailed description, with each claim standing on its own as a separate preferred embodiment of the invention.

[0114] Those skilled in the art will also appreciate that the various illustrative logic blocks, modules, circuits, and algorithmic steps described in conjunction with the embodiments herein can all be implemented as electronic hardware, computer software, or a combination thereof. In order to clearly illustrate the interchangeability between hardware and software, various illustrative components, blocks, modules, circuits, and steps are generally described above around their functions. Whether such functions are implemented as hardware or software depends on specific applications and the design constraints imposed on the entire system. A skilled person can implement the described functions in an alternative manner for each specific application, but such implementation decisions should not be interpreted as departing from the scope of protection of the present disclosure.

[0115] The steps of the method or algorithm described in conjunction with the embodiments herein may be directly embodied as hardware, a software module executed by a processor, or a combination thereof. The software module may be located in a RAM memory, a flash memory, a ROM memory, an EPROM memory, an EEPROM memory, a register, a hard disk, a mobile disk, a CD-ROM, or any other form of storage medium well known in the art. An exemplary storage medium is connected to the processor so that the processor can read information from the storage medium and can write information to the storage medium. Of course, the storage medium may also be an integral part of the processor. The processor and the storage medium may be located in an ASIC. The ASIC may be located in a user terminal. Of course, the processor and the storage medium may also be present in a user terminal as discrete components.

[0116] For software implementation, the techniques described in this application can be implemented with modules (e.g., procedures, functions, etc.) that perform the functions described in this application. These software codes can be stored in a memory unit and executed by a processor. The memory unit can be implemented within the processor or outside the processor. In the latter case, it is coupled to the processor in a communication manner via various means, which are well known in the art.

[0117] The above description includes examples of one or more embodiments. Of course, it is impossible to describe all possible combinations of components or methods for the purpose of describing the above embodiments, but it should be recognized by those skilled in the art that the various embodiments may be further combined and arranged. Therefore, the embodiments described herein are intended to cover all such changes, modifications and variations that fall within the scope of protection of the appended claims. In addition, with respect to the term "comprising" used in the specification or claims, the word is covered in a manner similar to the term "including", just as "including," is explained as a transitional word in the claims. In addition, any term "or" used in the specification of the claims is intended to mean "non-exclusive or".

Claims

1. A multi-target tracking method based on multi-sensor probability hypothesis density, characterized in that: include: S100. Establishing a tracking model for multi-sensor probability hypothesis density; S200. Perform state recursion on the multi-sensor probability hypothesis density according to the tracking model; S300. Obtaining the predicted strength of the multi-sensor probability hypothesis density; S400. Obtaining the posterior strength of the multi-sensor probability hypothesis density; S500. Prune and merge the probability hypothesis density of multiple sensors to estimate the target state and deviation.

2. A multi-target tracking method based on multi-sensor probability hypothesis density as claimed in claim 1, characterized in that: In S100, a tracking model is established for the multi-sensor probability hypothesis density, wherein the tracking model includes a linear target motion model and a measurement model, wherein the linear target motion model and the measurement model are respectively set as: x k =F k-1 x k-1 +w k-1 , l=1,2,...,L, where and denote the target state and the lth sensor measurement respectively; L is the number of sensors; F k-1 and is the state transfer matrix and measurement matrix; w k-1 and They are the covariance matrices and Zero-mean white Gaussian process noise and measurement noise; is the translation measurement deviation of the lth sensor.

3. The multi-target tracking method of multi-sensor probability hypothesis density as claimed in claim 1, characterized in that: In S200, state recursion is performed on the multi-sensor probability hypothesis density according to the tracking model. The specific method includes: using PHD recursion as the optimal multi-objective Bayesian recursion, by enhancing the state vector in Remember v k-1k-1 (x k-1 ,b k-1 ) is the posterior strength function at time k-1, then the predicted strength function v is derived kk-1 (x k ,b k ), assuming that the registration errors of different sensors are independent, that is, Said is the transfer strength of the lth sensor registration error, and the registration error It can be described as a first-order Gaussian Markov process, that is, After all sensor measurement data are obtained, the posterior strength v at time k can be obtained by sequential processing. kk .

4. A multi-target tracking method based on multi-sensor probability hypothesis density as claimed in claim 3, characterized in that: The calculation formula of the prediction strength function is: v kk-1 (x k ,b k )=∫[p s f x (x k |x k-1 )f(b k |b k-1 )+β kk-1 (x k ,b k |x k-1 ,b k-1 )]·v k-1k-1 (x k-1 ,b k-1 )dx k- 1db k-1 +γ k (x k ,b k ) Where p s is the survival probability of the existing target, f x and f b are the single target state transfer function and the registration error transfer density respectively; β kk-1 and γ k They represent the intensity function of the derived target random finite set and the intensity function of the outside world entering the new target random finite set respectively.

5. The multi-target tracking method of multi-sensor probability hypothesis density as claimed in claim 1, characterized in that: In S300, the prediction strength of the multi-sensor probability hypothesis density is obtained, and the specific method includes: Assume that the random finite set PHD and the derived random finite set PHD of the new target are expressed in the form of Gaussian mixture, that is: In the formula, J γ,k , as well as They are respectively the number of freshmen, freshmen weight, freshmen mean, freshmen deviation, deviation covariance, and freshmen covariance that determine the random finite set PHD of the new target from the outside world; J β,k , as well as To determine the number of derivatives, derivative weights, derivative transfer functions, derivative noise, derivative deviation transfer functions, and derivative deviation covariance of the random finite set PHD of the derived new target; Assume that the posterior strength function v at time k-1 is k-1k-1 (x,b) is in the form of a Gaussian mixture, that is: Where: J is the number of Gaussian terms; w is the weight of the jth Gaussian term, then the predicted intensity function v kk-1 (x k ,b k )for v kk-1 (x k ,b k )=v s,kk-1 (x k ,b k )+v β,kk-1 (x k ,b k )+v γ,kk-1 (x k ,b k ) In the formula, v s,kk-1 (x k ,b k ) is the survival strength, v β,kk-1 (x k ,b k ) is the derived strength, v γ,kk-1 (x k ,b k ) is the new strength.

6. A multi-target tracking method based on multi-sensor probability hypothesis density as claimed in claim 5, characterized in that: The posterior strength function at time k is specifically: In the formula, in, is the detection probability of the lth sensor; h l (z k |x x ,b k ) is the single target measurement likelihood function of the lth sensor; is the intensity function of a random finite set of clutter.

7. The multi-target tracking method of multi-sensor probability hypothesis density as claimed in claim 1, characterized in that: In S400, the posterior strength of the multi-sensor probability hypothesis density is obtained, and the specific method includes: The prediction strength function is recorded as: Then the update intensity function after multi-sensor sequential processing is:

8. A multi-target tracking method based on multi-sensor probability hypothesis density as claimed in claim 7, characterized in that: The Kalman prediction process at time k is as follows:

9. A multi-target tracking method based on multi-sensor probability hypothesis density as claimed in claim 8, characterized in that: The Kalman update process at time k is as follows: In the formula is the likelihood function of sensor l, is the new information of sensor l, is the predicted position of sensor l, is the measurement covariance of sensor l, is the measurement transfer function of sensor l.

10. An electronic device, characterized in that: include: one or more processors; A memory for storing one or more programs; When the one or more programs are executed by the one or more processors, the one or more processors implement any one of the multi-target tracking methods in claims 1-9.

Images