Label particle multi-extension target tracking method based on ellipse RHM

A multi-expansion target and particle number technology, which is applied in the field of tag particle multi-expansion target tracking based on elliptical RHM, can solve problems such as inability to achieve track tracking, and achieve the effect of fast calculation speed and high tracking accuracy.

Active Publication Date: 2021-12-28
XIDIAN UNIV
3 Cites 0 Cited by

AI-Extracted Technical Summary

Problems solved by technology

For the extended target, it can be modeled as an elliptical shape with the elliptical random hypersurface model (Random H...
the structure of the environmentally friendly knitted fabric provided by the present invention; figure 2 Flow chart of the yarn wrapping machine for environmentally friendly knitted fabrics and storage devices; image 3 Is the parameter map of the yarn covering machine
View more

Abstract

The invention relates to a label particle multi-extension target tracking method based on an ellipse RHM, and the method comprises the steps: building an augmentation space, and initializing a label particle set; carrying out measurement division, supplementing new label particles according to a measurement division unit and the measurement in the measurement division unit, and combining the new label particles with a survival label particle set; predicting and updating the label particle set generated by merging according to a prediction equation and a likelihood equation; performing label processing on the updated label particle set to obtain multiple extended target number estimation, and associating multiple extended target state estimation with a track; and after the label particle subset after label processing is resampled, carrying out target mass center state extension and shape estimation at the next moment. The invention is high in operation speed and high in tracking precision, the mass center state and shape of the multiple extended targets can be accurately estimated, the target track can be obtained, and different targets can be distinguished.

Application Domain

Technology Topic

Image

  • Label particle multi-extension target tracking method based on ellipse RHM
  • Label particle multi-extension target tracking method based on ellipse RHM
  • Label particle multi-extension target tracking method based on ellipse RHM

Examples

  • Experimental program(2)

Example Embodiment

[0099] Example one
[0100] Please refer to combined figure 2 and image 3 , figure 2 It is a block diagram of a process based on the label particle elliptical extension RHM multi target tracking method according to an embodiment of the present invention; image 3 Process is one kind of particles based on the label provided by the elliptical extension RHM multi target tracking method of the present invention is implemented. FIG. , Based on the tag elliptical particles of RHM Extensibility present embodiment shown in FIG target tracking method, comprising:
[0101] S1: In the initial time, the target allocation for each L 0 Particle labels, the initial state of the target sample, the label is initialized set of particles;
[0102] Initialization label particle set as follows:
[0103]
[0104]
[0105] in, I represents a state vector of the initial time of particles, I represents the weight of the particles of the initial time value, Tag indicates the i-th particle particle initial time, Θ 0 It represents the initial label set time, θ n It represents a target identification, N 0 It represents the number of targets present in the initial time;
[0106] In the present embodiment, the state of the target sample comprising the target motion parameters centroid [m xk m yk v xk v yk ] T And shape parameters [a k , B k , Φ k ] Meanwhile random sampling, wherein
[0107] [M xk m yk v xk v yk ] T Centriological motion parameters, m xk , M yk The coordinate of the K timeline in the X, Y axis, respectively, V xk , V yk The speed at the X, Y axis direction, respectively, A k , B k The long axis and short axis of the ellipse of the k timer, φ, respectively k Indicates the angle between the elliptical long axis and the X axis.
[0108] S2: According to the sessage label particle set at the previous time, the survival label particle set is obtained, and the measured measurement of the target is divided, and the M division unit is obtained, and each of each divided unit is measured to measure M new life label. Particles, resulting in new label particle set, wherein the same label is assigned to particles in the same dividing unit, and different labels are assigned different labels for particles in different divided cells;
[0109] Specifically, step S2 includes:
[0110] 2A. According to the sessage label particle set at the previous time, a set of survival label particles is obtained. In this embodiment, it is assumed that the survival label particle set in K-1 is:
[0111]
[0112] in, The state of the status of the ith living particles is shown in K-1, Indicates the weight of the ith living particles in K-1, A label representing the ith living particles in K-1 time, Label set, L k-1 Represents the number of survival tag particles in K-1, N k-1 Indicates the number of target estimates at the time of K-1;
[0113] 2b. Suppose the K is time to divide MS (Mean Shift, mean drift), get M division unit W j , j = 1, ... M, then the new life label set is
[0114] 2c. Make each of each component measurement sample to obtain M new ligament particles, obtain the number of new label particles And each new label particle weight in,
[0115]
[0116]
[0117] Where P b (•) Represents the new life target strength, in this embodiment, the new goal intensity is set to 0.01;
[0118] 2D. Number of new labels And each new label particle weight Get new label particles set:
[0119]
[0120] in, Streaming of the status of the i new part of the k hour, Indicates the weight of the i-th new particles in the k hour, A particle label of the i-th new particles in the k time.
[0121] S3: Combine the new label particle set and the survival label particle set to obtain a new label particle set;
[0122] Specifically, including, the new label particle set and the survival label particles are set, and the new label particle set is:
[0123]
[0124] The number of label particles for new label particles is:
[0125]
[0126] The label set of the new label particle set is:
[0127]
[0128] S4: The new label particle set is predicted and updated based on the predictive equation and the likelihood function, and the updated label particle set, wherein the number of particles and its tag remain unchanged during prediction and update;
[0129] Specifically, step S4 includes:
[0130] 4A. The status prediction of the new label particle set is predicted according to the prediction equation, and the label particle set after the prediction process is obtained, wherein
[0131] The forecast equation is:
[0132]
[0133]
[0134] During the prediction process, the number of particles and its tag remain unchanged:
[0135]
[0136] Where f represents the status transfer matrix, P S,k (·) Indicates the target survival probability, its value is a preset constant, V k Represents a covariance q k State noise, Represents the number of label particles for new label particles;
[0137] 4b. According to the current measurement and the likelihood function, the predictive label particle set is updated to get the updated label particle set. Among them, during the update process, the number of particles and its tag remain unchanged, ie,
[0138] Specifically, the label particle set after the current measurement and the likelihood function is updated, and the specific description is specifically described below:
[0139] Single expansion target quantitative function is:
[0140]
[0141] Where Z k Indicates the measurement, the proportional factor S obeys the mean Covariance Gauss distribution:
[0142]
[0143] The equation (18) can be written as:
[0144]
[0145] The closed form of the formula (20) under the frame of the ellipse RHM is:
[0146]
[0147] in,
[0148]
[0149]
[0150]
[0151]
[0152]
[0153]
[0154] b k = Z k -M k (28),
[0155] θ k = Arctan (z ky -M ky ,z kx -M kx ) (29),
[0156] Among them, in the formula, a k , B k φ k Indicate an elliptical shape parameter, θ k Indicates the measurement of the centroid x Mixed angle, Z k Representation measure, M k Indicate elliptical, Z ky , M ky The Y-axis coordinate of the measurement and centroid, respectively, Z kx , M kx Representation measurement and centroid respectively x Axis coordinate.
[0157] In order to facilitate calculation, the likelihood function represented by the formula (21) is in the form of logarithmic form:
[0158]
[0159] Suppose K time measurement is set M k Indicates the number of measurements, since the measurement is independent of each other, so the multi-expansion target measurement function is:
[0160]
[0161] Its logarithmic form is:
[0162]
[0163] Update to multi-objective prediction PHD and quantity test pseudo-like function by updating the multi-target prediction function
[0164] Di k (x | z k ) = L Z (x) d k|k-1 (x | z k-1 ) (33),
[0165] Where D k|k-1 (x | z k-1 ) Represents multi-objective prediction PHD.
[0166] S5: According to different target identities, the updated label particle set is divided into several subsets, obtain particle weights of each subset, and compare the particle weight and the preset threshold, according to the comparison result, obtained Label processing label particle set;
[0167] Specifically, step S5 includes:
[0168] 5A. Identification according to different targets θ n Classify the updated label particle set into several subsets:
[0169]
[0170] in, Indicates the updated label particle concentration target identification is θ n Particle number;
[0171]5b. Fitting a posterior extended target intensity for each subset, posteriori extended target intensity to obtain weight of each particle subsets and
[0172]
[0173]
[0174] That is, when the particle label hour, otherwise,
[0175] 5c. The weights and particle Compared with a preset threshold η, if The extended target survival; if , The disappearance of the extended object, removed from the concentrate particles subset of the updated tag, while the concentration of the target tag identification θ n Remove, after processing the tag label set of particles, wherein
[0176] After the tag label set of particles is treated:
[0177]
[0178] Among them, L k Tag represents the number of particles after the labeling process;
[0179] The number of k extended target time is:
[0180]
[0181] k extended target time set for the label:
[0182]
[0183] S6: Label set of particles through the labeling process, extended acquisition state based on the estimated target set, according to the state estimation extended target set to obtain a multi-extended target track;
[0184] In particular, steps S6 comprising:
[0185] 6a. The state of each particle through the tag label set of particles of the subset of the labeling process of the weighted average of the extended target state estimation result,
[0186] Extended target state estimation results:
[0187]
[0188] 6b. The extended target state estimation result obtained expanded state estimation target set
[0189]
[0190] 6c. In the filtering process, depending on the state estimation extended target set, having the same target identifier estimation result connected together represent different target identified by different icons or colors to give Extensibility target track.
[0191] Due to the expansion of the target state estimation concentrate containing a target identifier, and therefore different targets may be distinguished with the corresponding target identification.
[0192] S7: performs resampling set of particles through the label labeling process, the label obtained resampling set of particles.
[0193] Specifically, Step S7 comprising:
[0194] Each subset of the set of particles through the tag label processing, resampling set of particles obtained resampling Tags:
[0195]
[0196] In step S5, the label set of particles obtained after the labeling process is:
[0197]
[0198] right Resampling set of particles obtained:
[0199]
[0200] The formula (43), to give resampling label set of particles.
[0201] In the present embodiment, the process as resampling, Copying weight larger particles in the respective sub-tags from the target identifier to distinguish the particle, the smaller the particle weights discarded. Resampling survival tag label set of particles as the particles set in step S2.
[0202] RHM particles based on the tag oval extended multi-target tracking method of the present embodiment, based on the RFS, modeling a target shape of an ellipse extended random hypersurface model, based on new multi-objective nascent extended tag complement the particle method proposed Extensibility object tracking method based on the tag ellipse RHM particles, the process operation speed, high precision tracking, not only can accurately estimate the multi-state extended target centroid and shape can also be obtained target track, to distinguish different targets.

Example Embodiment

[0203] Example 2
[0204] MATLAB simulation of the present embodiment in conjunction with the embodiments based on the tag RHM particles elliptical extension of a multi target tracking method for tracking multiple extended effect will be described.
[0205] Simulation in the present embodiment, the tracking target is the target of uniform linear motion. A two-dimensional planar area of ​​the observed scene, the size of [-600,600] × [-600,600] (m 2 ), And to increase the random noise in the observation area as to simulate the real scene, the initial design of the position, size, direction of movement of the plurality of different target extended oval uniform linear motion in the observation area. Each extended target in the initial state, and the survival time of the shape parameters as shown in Table 1.
[0206] Table 1 initial state of each of the extended object shape parameter and survival time
[0207]
[0208] As can be seen from Table 1, the time from an initial extended target 1 starts to move, the target 2 appears when the extended 5s, the first 15s, and an extended target intersection extended target 2, the first 20s, certain expansion occurs in the simulation scenario 3, At the same time, the target 4 by an extension extended dividing to produce the target 2 by the first 30s, extended target 3 is incorporated in an extended target.
[0209] All the observation area are extended target model CV model, 15 extend certain Poisson's ratio, and the target amount of the measured uniformly distributed within the target range of diffusion, Poisson noise was 5, uniformly distributed in the observation field. Distribution scale factor is The sampling time interval T s = 1s, target survival probability P S = 0.99, P probability target newborn b = 0.01 detection probability of P D = 0.99, OSPA distance parameter c = 100, p = 2. The ET-LP-PHD filter algorithm, the initial timing for each target with L 0 = 600 sample particles, particles newborn set M = 60. Extended target state vector comprising a seven-dimensional vector of the motion and shape parameters, the state equation and the measurement equation, respectively:
[0210] x k = F k x k-1 + V k (45),
[0211] z k = Hx k + w k (46),
[0212] Among them, f k Shows a state transition matrix, v k Represents a covariance q k State noise, H is the measurement matrix, w k R represents covariance k The measurement noise, the parameter values ​​are as follows:
[0213]
[0214]
[0215]
[0216] Qi 1 = 0.1, q 2 = 0.5, R k = 0.5 2 I 2 ,
[0217] Which, I n Is an n × n unit matrix, It represents a direct product of two matrices.
[0218] Please refer to combined Figure 4 and Figure 5 , Figure 4 The target is extended on a simulation experiment provided a centroid trajectory and the actual shape of a schematic embodiment of the present invention, Figure 5 FIG extended target is measured on a simulation experiment according to an embodiment of the present invention. Figure 4 The "*" denotes an extended target centroid, oval shape represents the extended target real expansion. Figure 5 In "×" represents the measurement target generated each extension.
[0219] See Image 6 , Image 6 ET-LP-PHD is an embodiment of the present invention, a schematic diagram of the results of estimation. As shown, different extensions centroid target estimate represented by different icons, specifically: 1-14s spread estimation centroid and shape of the target 1 are represented by "·" denotes elliptical and, when the two extended target intersection 15s, extended target 2 is the extended target 1 undetected detected, 16S extended target 1 is detected, its label changes, corresponding with "o" and the oval represents the center of mass and the expanded shape, "×" and the oval represents the extended target estimated centroid 2 and shape, "*" indicates expansion and elliptical shape of the target and the estimated centroid 3, "+" denotes an extended ellipse and the target centroid and the estimated shape 4. Thus it can be seen, ET-LP-PHD extended target centroid in obtaining, on the basis of the estimated shape, can distinguish between different target, the target track is formed.
[0220] See Figure 7 , Figure 7 Is the estimated average number of target expanded view of a simulation according to an embodiment of the present invention. FIG 100 as the average number of Monte Carlo Simulation extended target estimate can be seen from the figure, the first 15s, since the two intersecting target, estimated leak situation occurs, the first 20s, and simultaneous targets derived newborn target, since a measurement derived a certain time near the front, it can be derived to predict a target. The ET-LP-PHD algorithm, when the weight of the particle child with the same label set of heavy and larger than the threshold η, it is considered a target survival or target newborn, these label particles subsets particle weight obtained by summing the number of target estimate, thus ET -LP-PHD filtering algorithm to estimate the number of target value slightly less than the real number, but still be able to accurately estimate the extension number of the target.
[0221] Please refer to combined Figure 8 and Figure 9 , Figure 8 An object of the present invention is the embodiment of the centroid average distance OSPA extended embodiment provides a schematic view, Figure 9 Is a schematic view of a major axis, and minor axis distances OSPA an extended toward a target angle estimate elliptical shape according to an embodiment of the present invention. from Figure 8As can be seen in the present invention, the method of the present invention can accurately estimate the expansion target priority.Due to the emergence of new expansion targets, the OSPA distance occurs because the expansion target cross has a peak, and the OSPA distance appears.from Figure 9 It can be seen that the ET-LP-PHD filtering algorithm based on new expansion target new label particles can be well estimated to scale the target shape.
the structure of the environmentally friendly knitted fabric provided by the present invention; figure 2 Flow chart of the yarn wrapping machine for environmentally friendly knitted fabrics and storage devices; image 3 Is the parameter map of the yarn covering machine
Login to view more

PUM

no PUM

Description & Claims & Application Information

We can also present the details of the Description, Claims and Application information to help users get a comprehensive understanding of the technical details of the patent, such as background art, summary of invention, brief description of drawings, description of embodiments, and other original content. On the other hand, users can also determine the specific scope of protection of the technology through the list of claims; as well as understand the changes in the life cycle of the technology with the presentation of the patent timeline. Login to view more.
the structure of the environmentally friendly knitted fabric provided by the present invention; figure 2 Flow chart of the yarn wrapping machine for environmentally friendly knitted fabrics and storage devices; image 3 Is the parameter map of the yarn covering machine
Login to view more

Similar technology patents

Farmland machine leading line extraction algorithm based on machine vision

ActiveCN103914836AFast operationImprove adaptability and poor precisionImage enhancementImage analysisUF - UltrafiltrationCorrosion
Owner:ZHENGZHOU UNIVERSITY OF LIGHT INDUSTRY

Classification and recommendation of technical efficacy words

  • Fast operation
  • Improve tracking accuracy

Continuous casting breakout prediction method based on neural network

InactiveCN101850410AFast operationImprove recognition effect and forecast accuracyPre treatmentPrediction rate
Owner:PANGANG GROUP VANADIUM TITANIUM & RESOURCES +3
Who we serve
  • R&D Engineer
  • R&D Manager
  • IP Professional
Why Eureka
  • Industry Leading Data Capabilities
  • Powerful AI technology
  • Patent DNA Extraction
Social media
Try Eureka
PatSnap group products