CN109581353B - Multi-target tracking method and system based on automobile radar - Google Patents
Multi-target tracking method and system based on automobile radar Download PDFInfo
- Publication number
- CN109581353B CN109581353B CN201811423573.XA CN201811423573A CN109581353B CN 109581353 B CN109581353 B CN 109581353B CN 201811423573 A CN201811423573 A CN 201811423573A CN 109581353 B CN109581353 B CN 109581353B
- Authority
- CN
- China
- Prior art keywords
- target
- covariance matrix
- track
- flight path
- state
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Active
Links
Images
Classifications
-
- G—PHYSICS
- G01—MEASURING; TESTING
- G01S—RADIO DIRECTION-FINDING; RADIO NAVIGATION; DETERMINING DISTANCE OR VELOCITY BY USE OF RADIO WAVES; LOCATING OR PRESENCE-DETECTING BY USE OF THE REFLECTION OR RERADIATION OF RADIO WAVES; ANALOGOUS ARRANGEMENTS USING OTHER WAVES
- G01S13/00—Systems using the reflection or reradiation of radio waves, e.g. radar systems; Analogous systems using reflection or reradiation of waves whose nature or wavelength is irrelevant or unspecified
- G01S13/66—Radar-tracking systems; Analogous systems
Abstract
The invention provides a multi-target tracking method and a system based on an automobile radar, which comprises the following steps: clustering each detection target by using density clustering to generate each effective target; respectively calculating the correlation degree of each effective target and each previous period flight path to generate a first covariance matrix; generating an evaluation matrix by utilizing dynamic alpha filtering according to the first covariance matrix and the second covariance matrix in the previous period; performing Hungary assignment according to the priorities and the evaluation matrix to generate matching tracks of the effective targets; performing Kalman filtering according to the effective target state, the flight path in the previous period and the second covariance matrix in the previous period to generate a second covariance matrix in the current period; and generating the flight path of each current period and an object set after resampling of the current period by utilizing Monte Carlo multiple probability sampling according to the state of each effective target, the corresponding matched flight path state of each effective target, the first covariance matrix and the second covariance matrix of the current period. The method and the device have the beneficial effect of considering both the nonlinear estimation precision and the real-time property.
Description
Technical Field
The invention relates to the technical field of automobile radar tracking, in particular to a multi-target tracking method and system based on an automobile radar.
Background
The radar is an indispensable component of the intelligent automobile and is used for target detection and target tracking. In an actual traffic environment, the environment where the automobile radar is located is complex. In order to meet the requirements of the automobile radar, effective targets need to be screened out from a large amount of target data, corresponding track management is established, and state information of target vehicles is accurately obtained. And carrying out multi-target tracking in real time to meet the requirement of early warning.
The automotive radar Tracking algorithm generally adopts a flight path correlation method such as Joint Probability Data Association (JPDA) and multi-Hypothesis Tracking (MHT) to combine with an improved algorithm of KF such as Kalman Filter (KF) and Extended Kalman Filter (EKF) and Unscented Kalman Filter (Unscented Kalman Filter, UKF). However, in actual testing, as the number of clutter and target echoes increases, the total number of combined events increases exponentially. On the other hand, due to the volatility of the radar signal, the non-linearity problem of the system needs to be considered in the radar maneuvering target tracking. The improved algorithm of the KF and the KF has good real-time performance, but the nonlinear estimation precision is lower.
Therefore, how to solve the problem of poor compatibility between the non-linear estimation accuracy and the real-time performance is a technical problem to be solved at present.
Disclosure of Invention
In order to overcome the defects in the prior art, the invention provides a multi-target tracking method and a multi-target tracking system based on an automobile radar, the method comprises the steps of decomposing track association into two steps, screening effective targets through a density clustering algorithm, and fusing echo targets from the same vehicle; secondly, setting a truncation distance through a Nearest Neighbor (NN) algorithm to limit the potential decision number, and enabling echoes obtained by primary screening of the truncation distance to become candidate echoes. And converting the association problem into an assignment problem to be processed, matching each effective target with each previous period flight path by using Hungarian assignment, obtaining a matched flight path corresponding to each effective target, and finishing flight path association. The target is tracked by utilizing Kalman filtering and Monte Carlo multivariate probability sampling, and the algorithm has the beneficial effect of better considering both nonlinear estimation precision and instantaneity.
In order to achieve the aim, the invention provides a multi-target tracking method based on an automobile radar, which comprises the following steps:
acquiring a plurality of detection targets corresponding to each vehicle to be detected in the current period, and clustering each detection target by using a density clustering algorithm to generate each effective target; the effective targets include: a valid target state;
calculating the correlation between each effective target and each acquired previous period flight path respectively to generate a first covariance matrix;
generating an evaluation matrix by utilizing dynamic alpha filtering according to the first covariance matrix and the acquired second covariance matrix in the previous period;
performing Hungary assignment according to the priority of each previous period flight path and the evaluation matrix to generate a matched flight path corresponding to each effective target; the matching track comprises: matching a track state;
kalman filtering is carried out according to each effective target state, each previous period flight path and the previous period second covariance matrix, and a current period second covariance matrix is generated;
and generating the current period flight path of each effective target and the object set after the current period resampling by utilizing Monte Carlo multivariate probability sampling according to the state of each effective target, the matched flight path state corresponding to each effective target, the first covariance matrix and the current period second covariance matrix.
The invention also provides a multi-target tracking system based on the automobile radar, which comprises:
the system comprises a clustering unit, a detection unit and a control unit, wherein the clustering unit is used for acquiring a plurality of detection targets corresponding to each vehicle to be detected in the current period and clustering each detection target by using a density clustering algorithm to generate each effective target; the effective targets include: a valid target state;
the correlation unit is used for calculating the correlation between each effective target and each acquired previous period flight path respectively to generate a first covariance matrix;
the dynamic filtering unit is used for generating an evaluation matrix by utilizing dynamic alpha filtering according to the first covariance matrix and the acquired second covariance matrix of the previous period;
the assignment unit is used for performing Hungarian assignment according to the priority of each previous period flight path and the evaluation matrix to generate a matched flight path corresponding to each effective target; the matching track comprises: matching a track state;
the Kalman filtering element is used for carrying out Kalman filtering according to each effective target state, each previous period flight path and the previous period second covariance matrix to generate a current period second covariance matrix;
and the sampling unit is used for generating the current period flight path of each effective target and the object set after the current period resampling by utilizing Monte Carlo multiple probability sampling according to the state of each effective target, the corresponding matched flight path state of each effective target, the first covariance matrix and the current period second covariance matrix.
The invention also provides computer equipment which comprises a memory, a processor and a computer program stored on the memory and capable of running on the processor, wherein the processor executes the program to realize the steps of the multi-target tracking method based on the automobile radar.
The present invention also provides a computer-readable storage medium having stored thereon a computer program which, when being executed by a processor, carries out the steps of the automotive radar-based multi-target tracking method.
The invention provides a multi-target tracking method and a system based on an automobile radar, which comprises the following steps: acquiring a plurality of detection targets corresponding to each vehicle to be detected in the current period, and clustering each detection target by using a density clustering algorithm to generate each effective target; the effective targets include: a valid target state; calculating the correlation degree of each effective target and each acquired previous period flight path respectively to generate a first covariance matrix; generating an evaluation matrix by utilizing dynamic alpha filtering according to the first covariance matrix and the acquired second covariance matrix of the previous period; performing Hungarian assignment according to the priority of each previous period flight path and the evaluation matrix to generate a matched flight path corresponding to each effective target; the matching track comprises: matching a track state; performing Kalman filtering according to each effective target state, each previous period flight path and the previous period second covariance matrix to generate a current period second covariance matrix; according to the effective target states, the matched flight path states corresponding to the effective targets, the first covariance matrix and the current period second covariance matrix, monte Carlo multiple probability sampling is utilized to generate the current period flight paths of the effective targets and the object set after current period resampling, multi-target tracking is achieved, and the method has the beneficial effect of taking the nonlinear estimation precision and real-time into consideration well.
Drawings
In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings used in the description of the embodiments or the prior art will be briefly described below, it is obvious that the drawings in the following description are only some embodiments of the present invention, and for those skilled in the art, other drawings can be obtained according to the drawings without creative efforts.
FIG. 1 is a flow chart of a multi-target tracking method based on automotive radar according to the present application;
FIG. 2 is a flow chart of a multi-target tracking method based on automotive radar in an embodiment of the present application;
FIG. 3 is a two-dimensional plot of velocity versus distance in an embodiment of the present application;
FIG. 4 is a flowchart of step S202 in an embodiment of the present application;
FIG. 5 is a flow chart of a tracking algorithm in an embodiment of the present application;
FIG. 6 is a flowchart of step S208 in an embodiment of the present application;
FIG. 7 (a), FIG. 7 (b) and FIG. 7 (c) are graphs of the tracking results of different linear scenes of a single target in an embodiment of the present application;
FIG. 8 (a), FIG. 8 (b) and FIG. 8 (c) are graphs of single target tracking results of different lane-change scenarios in an embodiment of the present application;
FIG. 9 (a), FIG. 9 (b) and FIG. 9 (c) are graphs of single target tracking results of different curve scenes in an embodiment of the present application;
FIGS. 10 (a) and 10 (b) are graphs of multi-target tracking results of different test scenarios in an embodiment of the present application;
fig. 11 is a schematic structural diagram of a multi-target tracking system based on an automotive radar according to the present application.
Detailed Description
The technical solutions in the embodiments of the present invention will be clearly and completely described below with reference to the drawings in the embodiments of the present invention, and it is obvious that the described embodiments are only a part of the embodiments of the present invention, and not all of the embodiments. All other embodiments, which can be derived by a person skilled in the art from the embodiments given herein without making any creative effort, shall fall within the protection scope of the present invention.
The terms "first," "second," "8230," "8230," and the like as used herein do not particularly denote any order or sequence, nor are they intended to limit the invention, but rather are used to distinguish one element from another or from another element described in the same technical term.
As used in this application, the terms "comprising," "including," "having," "containing," and the like are open-ended terms that mean including, but not limited to.
As used herein, "and/or" includes any and all combinations of the stated items.
Aiming at the defects in the prior art, the invention provides a multi-target tracking method based on an automobile radar, a flow chart of which is shown in figure 1, and the method comprises the following steps:
s101: and acquiring a plurality of detection targets corresponding to each vehicle to be detected in the current period, and clustering each detection target by using a density clustering algorithm to generate each effective target. Wherein the effective targets include: a valid target state.
S102: and calculating the correlation degree of each effective target and each acquired previous period flight path respectively to generate a first covariance matrix.
S103: and generating an evaluation matrix by utilizing dynamic alpha filtering according to the first covariance matrix and the acquired second covariance matrix in the previous period.
S104: and performing Hungarian assignment according to the priority and the evaluation matrix of each previous period of flight path to generate a matched flight path corresponding to each effective target. Wherein, the track of matching includes: and matching the track state.
S105: and performing Kalman filtering according to each effective target state, each previous period flight path and the previous period second covariance matrix to generate a current period second covariance matrix.
S106: and generating the current period flight path of each effective target and the object set after resampling in the current period by utilizing Monte Carlo multivariate probability sampling according to the state of each effective target, the corresponding matched flight path state of each effective target, the first covariance matrix and the second covariance matrix in the current period.
As can be seen from the flow shown in fig. 1, the present application decomposes the track association into two steps, first, selects an effective target of each class from each detected target of each class by a density clustering algorithm, and fuses the detected targets from the vehicles to be detected; secondly, matching each effective target with each previous cycle track by using Hungary assignment to obtain a matched track corresponding to each effective target, and finishing track association; and filtering by taking the effective target state as measurement and the track state as target estimation. The method has the advantages that the characteristic of small Kalman filtering calculation amount is utilized, the algorithm is optimized in a Kalman filtering pre-estimation mode, the updated flight path of the current period is generated by Monte Carlo multi-element probability sampling, real-time multi-target tracking is carried out, and the method has the beneficial effect of considering both nonlinear estimation precision and instantaneity.
In order to make those skilled in the art better understand the present invention, a more detailed embodiment is listed below, and as shown in fig. 2, the multi-target tracking method based on automotive radar provided by the embodiment of the present invention includes the following steps:
s201: and acquiring a plurality of detection targets corresponding to each vehicle to be detected in the current period.
When the system is specifically implemented, the automobile radar system is mainly used for detecting, tracking and alarming the targets in front of the vehicle, and mainly comprises: cameras, radars, and alarms. The automobile radar system can detect vehicles within +/-10 degrees in front and track detection targets of the vehicles in real time. In an actual road environment, it is inevitable that the same vehicle has a plurality of strong reflection centers, and each vehicle is detected as a plurality of detection targets.
The automobile radar system in this embodiment is specifically a forward automobile anti-collision radar, the forward automobile anti-collision radar adopts a Frequency Modulated Continuous Wave (FMCW) system, the transmitted Wave is a high-Frequency Continuous Wave, and the Frequency of the high-Frequency Continuous Wave changes with time according to a sawtooth Wave rule. The change rule of the frequency of the echo received by the radar is the same as the change rule of the frequency of the echo transmitted by the radar, the detection distance R of the radar reaching a detection target can be calculated by utilizing the time difference, the included angle between the detection target and the normal direction of the radar, namely the detection angle theta, is measured through the design of four antennas, the calculation formula of the specific detection distance R is shown as a formula (1), and the calculation formula of the detection angle theta is shown as a formula (2):
wherein, the detection distance R is the target distance from the radar to the detection target, c is the speed of light, T is the period of the radar emission signal, Δ f is the difference frequency between the radar emission signal and the radar receiving signal, B is the signal bandwidth of the radar emission signal, the detection angle theta is the included angle between the detection target and the radar normal direction, and λ is the wavelength of the electromagnetic wave emitted by the radar antenna,
for phase differences between radar antennas, d i Is the distance between the radar antennas.
Two-dimensional fast fourier transform is performed on the data acquired by the radar in the period, so that a two-dimensional speed-distance graph can be obtained, as shown in fig. 3. And detecting by a Constant False-Alarm Rate (CFAR) to obtain each detection target and target detection data corresponding to each detection target in the period. Wherein the target detection data comprises: and detecting the speed V, the detection distance R, the detection angle theta and the target energy, and calculating the target acceleration through the speed. The invention is not limited thereto.
Simultaneously acquiring each updated track generated in the previous period as each previous period track S of the period i And an initial covariance matrix. Wherein the previous cycle track comprises: the number of the track cycles, the accumulated error value and the track state corresponding to the previous cycle of the track (referred to as the previous cycle of the track state for short), etc. And if the number of the track cycles is large, the track of the previous cycle is considered to be more stable. Accumulating error values for identifying flight paths from the previous cycleUp to now, the accumulation of the covariance when the previous period flight path matches the valid target is used to evaluate the degree of engagement between the valid target and the previous period flight path.
S202: and clustering the detection targets by using a density clustering algorithm to generate effective targets. Wherein the valid targets include: a valid target state.
The method and the device have the advantages that the density peak clustering Algorithm (dense peak clustering Algorithm) is utilized to fuse detection targets from the same vehicle into an effective target, and the stability and the confidence degree of track association are improved. Setting the data set of the detection target to X = { X = { [ X ] (1) ，x (2) ，…，x (i) ，…，x (m) ；x (i) ∈R n In which x (i) Denotes the ith detection target, x (i) For an n-dimensional vector, m represents the number of detection targets in the data set of detection targets.
In specific implementation, as shown in fig. 4, step S202 specifically includes the following steps:
s301: and generating the local density of each detection target and the target distance of each detection target by using a local density algorithm according to the target detection data of each detection target and a preset truncation distance.
When the method is implemented, firstly, the local density and the target distance are calculated, and the truncation distance d is set c Respectively calculating the local density of the detection target, namely judging the distance d between the ith detection target and each of the other m-1 detection targets ij Whether at the truncation distance d c If yes, increasing the local density of the ith detection target by 1; if not, the local density of the ith detection target is unchanged, wherein the value range of each local density is 0 to (m-1).
Local density ρ of each detection target i The specific calculation formula is shown as formula (3):
wherein ρ i To detect an object x (i) The local density of (a) is,χ (x) is a decision function as shown in equation (4):
when d is ij -d c When less than 0, x (x) is 1; when d is ij -d c When x is more than or equal to 0, x (x) is 0.
Wherein, the target distance d between each detection target and other detection targets ij The specific calculation process of (2) is shown in formula (5):
d ij ＝f(R i ,R j ,V i ,V j ,θ i ,θ j ) (5)
wherein, d ij Indicating a detected object x (i) And x (j) Distance between, x (j) Is divided by x (i) M-1 of (1), d c To cut off the distance, V i Is x (i) Detection speed of R i Is x (i) Is detected by a detection distance theta i Is x (i) Angle of detection of V j Is x (j) Detection speed of R j Is x (j) Detected distance of (a), theta j Is x (j) The angle of detection of (1).
S302: and classifying the detection targets according to the local densities to generate a clustering center of each type of detection target.
In specific implementation, all the detection targets are classified according to the local densities, and the clustering distances of all the detection targets are generated.
Specifically, the detection target x satisfying the formula (6) (i) And x (j) The classification into one category:
i.e. the local density is more than or equal to the detection target x (i) Local density of (p) i And a target distance d ij Is less than or equal to the truncation distance d c Detection target x of (j) And dividing the detected objects into a category, and taking the detected object with the largest local density in each category of detected objects as a clustering center.
The clustering distance calculation process of various detection targets is as follows: to detect the target x (i) Cluster distance δ of the associated class i The calculation process is taken as an example, and the detection target x is assumed (i) Local density of (p) i To 3, the detection targets x with the other local densities of 3 or more are calculated respectively (j) Target distance d therebetween ij And the minimum target pitch mind is set ij Distance d from the cut-off c Making a comparison if mind ij ≤d c Then mind will be ij As detection target x (i) Cluster distance δ of i The specific calculation formula is shown as formula (7):
wherein the target ρ is detected j To detect an object x (j) The local density of (a).
S303: and weighting each type of detection target according to each clustering center to generate an effective target corresponding to each type of detection target.
And carrying out intra-class weighting according to the clustering center of each class of detection target, and carrying out weighting processing to generate the effective target of each class of detection target by taking the clustering center as a main part and other detection targets in the class as auxiliary parts.
S203: and calculating the correlation degree of each effective target and each acquired previous period flight path respectively to generate a first covariance matrix.
In specific implementation, according to a Neighbor-Neighbor (NN) idea, a correlation degree is calculated by using each effective target state and each acquired previous cycle track state, and a first covariance matrix is generated. Wherein the valid target states include: information such as distance, speed, angle, acceleration and energy, the track state includes: and the track information such as distance, speed, angle, acceleration, energy and the like. The first covariance matrix is formed by i x j Cov ij Composition of each Cov ij Is calculated byThe formula is shown in formula (8):
Cov ij ＝d(X i ,Z j ) (8)
wherein, X i Represents the ith valid target state; z j Representing the jth track state of the previous cycle. d (X) i ,Z j ) Function representation X i And Z j Multidimensional spatially weighted Euclidean distance, cov, for each state (distance, velocity, angle, acceleration, and energy) ij And representing the covariance of the ith effective target and the jth track in the previous period, and finally forming a covariance matrix of i x j.
S204: and generating an evaluation matrix by utilizing dynamic alpha filtering according to the first covariance matrix and the acquired second covariance matrix in the previous period.
In specific implementation, the α filtering is z = a × x + b × y, a and b represent weights, x is a first covariance matrix, and y is an acquired second covariance matrix (detected deviation) in a previous period. If the first covariance matrix (predicted deviation) is larger than the second covariance matrix in the previous period, the predicted deviation is considered to be larger, and the detected deviation in the previous period should be more confident, the weight of the detected deviation is increased in the next period, and the weight of the predicted deviation is reduced.
On the contrary, if the first covariance matrix (predicted deviation) is smaller than the second covariance matrix (detected deviation) in the previous cycle, the predicted deviation is considered to be small, and if the detected deviation in the previous cycle is more believed, the weight of the detected deviation is reduced in the next cycle, and the weight of the predicted deviation is increased.
According to the nearest neighbor NN algorithm, an evaluation matrix C is generated by utilizing dynamic alpha filtering according to a first covariance matrix and a second covariance matrix of each effective target tracking result, as shown in a formula (9):
wherein, c mn Representing the matching of the mth valid target with the nth previous cycle track and the covariance between the mth valid target and the nth previous cycle track.
S205: and calculating the corresponding priority of each previous period track according to the track cycle number of each previous period track and the accumulated error value.
In particular, degree of priority
Is shown in equation (10):
wherein i is the ith track in the previous cycle track, k is the current cycle value, the value range of k is a positive integer greater than or equal to 1,
the number of track cycles of the ith track of the previous track,
and accumulating error values for the flight path of the ith flight path of the previous period of flight path.
For example, when k =3, it means that the current period is the 3 rd period, and it is assumed that the number of track cycles of the 1 st track of the previous period (i.e., the 2 nd period) is T 3 1 =1, the number of track cycles of the 2 nd track of the previous cycle is T 3 2 ＝2。
S206: and performing Hungarian assignment according to the priority and the evaluation matrix of each previous period flight path to generate a matched flight path corresponding to each effective target.
The number of valid targets is typically not equal to the number of tracks generated in the previous cycle. To match m effective targets with n previous cycle tracks, similarity between the effective targets and the previous cycle tracks needs to be measured by constructing an evaluation matrix with the number of rows equal to the number of columns, and then the matched tracks are obtained. Wherein, the matching track includes: and matching the track state.
In specific implementation, the process of solving the optimal solution is converted into an assignment problem. The assignment problem can be described by a mathematical model, as shown in equations (11) and (12):
wherein, c ij Match ith target with jth track, x ij To evaluate the data of the ith row and j column of matrix C.
s.t. represents compliance (subject.to), i.e. satisfies the following formula:
after assignment, each element of the evaluation matrix C takes a value of 0 or 1, and the sum of each row and each column of the evaluation matrix C is 1:
z represents an assignment error, x ij =1 represents assignment. And assigning the effective target according to the evaluation matrix to obtain the minimum z, wherein the assignment error is minimum. The matching tracks are obtained by hungarian assignment. On the basis, a certain threshold value is set, and if the difference between the effective target and the previous cycle of the flight path exceeds the threshold value, matching is not carried out.
And in the truncation distance, generating a priority according to the track periodicity of each previous cycle track and the accumulated error value. And sequencing the priorities from high to low, and determining the previous cycle track with the highest priority (assumed as the 3 rd previous cycle track). Then, in the evaluation matrix, the covariance (c) with the previous cycle track (3 rd previous cycle track) is found m3 ) Minimum effective target (assuming c in the evaluation matrix) 53 Is c m3 The 5 th valid target and the 3 rd previous cycle flight pathMinimum covariance). If the covariance (c) of the previous cycle track (3 rd previous cycle track) m3 ) If the smallest valid target (i.e. the 5 th valid target) is within the cutoff distance, the valid target matches the previous cycle track (the 5 th valid target matches the 3 rd previous cycle track), i.e. the previous cycle track is the matching track of the valid target. If the covariance (c) of the previous cycle track (3 rd previous cycle track) m3 ) The smallest valid target (i.e., the 5 th valid target) is not within the cutoff distance, then the valid target does not match the previous cycle track (the 5 th valid target does not match the 3 rd previous cycle track). And (4) sequencing the tracks of the next previous period from high to low according to the priority, and by parity of reasoning, finally obtaining the matching relation between each previous period track and each effective target by combining Hungarian assignment, and generating the matched tracks of each effective target.
S207: and performing Kalman filtering according to each effective target state, each previous period flight path and the previous period second covariance matrix to generate the current period second covariance matrix.
In specific implementation, the effective target state is used as measurement, a flight path in a previous period is used as a motion state, a second covariance matrix in the previous period is used as a filtering covariance matrix, a noise matrix obtained by combining a physical model and an empirical value is combined, and an updated second covariance matrix, namely a second covariance matrix in the current period, is obtained according to a Kalman filtering algorithm.
S208: and generating the current period flight path of each effective target and the object set after resampling in the current period by utilizing Monte Carlo multivariate probability sampling according to the state of each effective target, the corresponding matched flight path state of each effective target, the first covariance matrix and the second covariance matrix in the current period.
When the radar starts working, the tracking algorithm randomly initializes the evaluation objects, obtains a plurality of evaluation objects (the evaluation object states comprise distance, speed, angle and acceleration), and forms an object set
And generating object set means
Sum variance
When each effective target starts to track, if no matched flight path exists, a new flight path is generated. The tracking algorithm flow is shown in FIG. 5, where KF 1 ～KF N Representing N times Kalman filtering, W 1 ～W N Representing the weights of the N objects.
Establishing a motion state model as shown in formula (13):
wherein x is k ＝f(x k-1 )+w k Is an equation of state, z k ＝h(x k )+v k For the measurement equation, x k Is the state matrix of the valid target at time k, z k An observation matrix of valid targets at time k, w k Is a state noise matrix, v k To measure the noise matrix. In this example x k Representing track status, z k Indicating a valid target state.
And updating the sampling object by using Kalman filtering, and predicting and updating the state to obtain a second covariance matrix in the current period.
As shown in fig. 6, the specific execution of step S208 includes the following steps:
s401: and performing weighted calculation according to the effective target states, the matched track states corresponding to the effective targets and the first covariance matrix to generate a sampling center.
In specific implementation, in the k period, the ith effective target state and the corresponding matched track state are weighted to obtain a sampling center
Wherein i represents the ith effective target, and k represents the current matched track corresponding to the ith effective targetA period value.
S402: and sampling by using a sampling function according to the obtained object set subjected to resampling in the previous period, the sampling center, the first covariance matrix and the second covariance matrix in the current period to generate a sampling object. Wherein the sampling object includes: object state and object covariance.
During specific implementation, a mixed covariance matrix between each effective target and the matched track is calculated according to the first covariance matrix and the second covariance matrix, and the mixed covariance matrix represents the multivariate association range between the matched tracks of the effective targets and the effective targets, and the multivariate association range existing between the current cycle track and the previous cycle track. And screening the objects in the multivariate relevance range by taking the sampling center as a base point through a sampling function to obtain an updated object set. Wherein the sampling function comprises: gaussian sampling function, poisson disk sampling function, etc.
k time object set
Representing a set of sampled objects that meet the sampling requirements,
which represents the center of the sample,
representing a range of multivariate associations, e.g. 100 objects before sampling and 50 objects after sampling, i.e. a set of objects meeting the sampling requirements
Has 50 objects as shown in formula (14):
s403: and performing weight calculation according to the object state and the object covariance to generate the expectation of each sampling object, and updating the current periodic flight path of each effective target according to the expectation of each sampling object.
In specific implementation, a formula for outputting the target state after sampling is shown in formulas (15) to (17), and the current period track of each effective target is generated:
obtaining a sampled set of objects
Then, the weights of the object set need to be calculated, and the calculation formula is shown as formula (15):
according to the weight of each object and the state of the object, weighting and summing to obtain a current track state estimation and a covariance estimation, and using the expectation of the state of the object as the current track state estimation, as shown in formula (16), the variance estimation is as shown in formula (17):
wherein, the first and the second end of the pipe are connected with each other,
the current periodic track state of the valid target at time k,
is the covariance of the current periodic track of the valid target at time k.
Outputting weighted track states
And according to the states, reflecting the track position in the form of transverse distance, transverse speed, longitudinal distance and longitudinal speed, namely the output of the periodAnd (4) outputting the updated track.
S404: and resampling by utilizing a preset sampling rule according to the sampling center to generate an object set resampled in the current period. Wherein the sampling center is the sampling center in S401
The preset sampling rule generation process is as shown in (18):
(R',V',N)＝f(R,V,SNR) (18)
as shown in formula (18), R and V represent boundary values of the distance-velocity model, which are used to divide the current effective target velocity and the distance fluctuation range, and if the range is exceeded, the current effective target velocity and the distance fluctuation range are regarded as not being the same effective target definitely; SNR represents the effective target signal-to-noise ratio state. R ', V' represent the resampling range in the distance dimension and the speed dimension, N represents the number of resampled objects, and R ', V' and N are numerical values generated according to an exponential relationship obtained along with an SNR change rule obtained by a Monte Carlo experiment.
In order to make the present invention better understood by those skilled in the art, the drive test data verification results of the present application are given below: in the embodiment, the test scene is a three-lane highway, and the echo data of the front vehicle is collected and the tracking result of the front vehicle is verified.
For effective target track association, performing multi-frame statistics on 4 different scenes respectively according to the algorithm association result and the actual corresponding relation, wherein a track association result statistical table is shown in table 1:
TABLE 1
And the unassociated representation indicates that the target cannot find a corresponding matched track. Effective correlation ratio = number of correct correlations/(number of correct correlations + number of incorrect correlations). The result of the track association of the algorithm is reflected in table 1, the average effective association rate is about 95% for 4 scenes, the target and the track can be effectively matched in a multi-target scene, and a better association effect is achieved.
In addition, for different numbers of targets, the method of the application compares the time performance and the association rate with JPDA and MHT, and the performance comparison of the association algorithm is shown in Table 2:
TABLE 2
The results in table 2 show that, for the actual drive test scene of the automobile radar, the difference between the effective correlation rates of the three correlation algorithms is not large, and the three correlation algorithms have higher effective correlation rates; but the method is obviously superior to JPDA and MHT in time performance, and has important practical significance in that good time performance is still maintained when the target number is increased.
The method and the device perform effect analysis on a plurality of different scenes, and mainly comprise a straight-line driving scene, a curve driving scene and a multi-target scene. The tracking tracks of the target in a certain period of time are recorded, the single-target tracking results are shown in fig. 7 (a), 7 (b), 7 (c), 8 (a), 8 (b), 8 (c), 9 (a), 9 (b) and 9 (c), and the multi-target tracking results are shown in fig. 10 (a) and 10 (b).
Fig. 7 (a), 7 (b), and 7 (c) show the tracking results of a single target in three straight line scenes, fig. 8 (a), 8 (b), and 8 (c) show the tracking results of a single target in three lane change scenes, and fig. 9 (a), 9 (b), and 9 (c) show the tracking results of a single target in three curve scenes. Since the reflection surface of the detection target is mainly horizontal, the tracking has a certain horizontal fluctuation. As shown in fig. 7 (a), 7 (b), and 7 (c), the vehicle tracking result has a certain degree of curvature in the straight-driving scene; as shown in fig. 8 (a), 8 (b), 8 (c), 9 (a), 9 (b) and 9 (c), the lane change scene and the curve scene are not obvious because the vehicle has a transverse displacement, and the track transverse fluctuation is not obvious. In the three scenes, the target track is continuous and stable, and the method has good scene adaptability.
Fig. 10 (a) and 10 (b) show the tracking results of multiple target different test scenarios. As shown in fig. 10 (a) and 10 (b), in a multi-target scene, the method and the system can well distinguish different targets, accurately show the motion state of each effective target, have stable track and good multi-target tracking capability, and provide necessary conditions for automobile safety early warning.
According to the multi-target tracking method and system based on the automobile radar, the traditional track association method is improved according to the requirements of the automobile radar on time performance and nonlinear estimation accuracy, track association is divided into two small links of clustering and assigning, monte Carlo multiple probability sampling is utilized, nonlinear estimation accuracy is improved, and a good multi-target tracking effect is obtained. The method provides good conditions for the development of intelligent automobiles and unmanned driving, is beneficial to reducing traffic accidents and the development of urban traffic intellectualization, and has important practical significance.
Based on the same application concept as the multi-target tracking method based on the automobile radar, the invention also provides a multi-target tracking system based on the automobile radar, which is described in the following embodiment. Because the principle of solving the problems of the multi-target tracking system based on the automobile radar is similar to that of the multi-target tracking method based on the automobile radar, the implementation of the multi-target tracking system based on the automobile radar can refer to the implementation of the multi-target tracking method based on the automobile radar, and repeated parts are not repeated.
Fig. 11 is a schematic structural diagram of a multi-target tracking system based on an automotive radar according to an embodiment of the present application, and as shown in fig. 11, the multi-target tracking system based on the automotive radar includes: clustering unit 101, correlation unit 102, dynamic filtering unit 103, assignment unit 104, kalman filtering unit 105, and sampling unit 106.
The clustering unit 101 is configured to obtain a plurality of detection targets corresponding to each vehicle to be detected in a current cycle, and cluster each detection target by using a density clustering algorithm to generate each effective target; the effective targets include: a valid target state.
And the correlation unit 102 is configured to perform correlation calculation on each of the effective targets and the acquired previous cycle flight path respectively to generate a first covariance matrix.
And the dynamic filtering unit 103 is configured to generate an evaluation matrix by using dynamic alpha filtering according to the first covariance matrix and the acquired second covariance matrix in the previous period.
And the assigning unit 104 is configured to perform hungarian assignment according to the priority of each previous cycle flight path and the evaluation matrix to generate a matched flight path corresponding to each valid target. Wherein, the matching track includes: and matching the track state.
And a kalman filtering unit 105, configured to perform kalman filtering according to each effective target state, each previous period flight path, and the previous period second covariance matrix, so as to generate a current period second covariance matrix.
And a sampling unit 106, configured to generate a current period flight path of each effective target and an object set after resampling in the current period by using monte carlo multivariate probability sampling according to each effective target state, a matching flight path state corresponding to each effective target, the first covariance matrix, and the second covariance matrix in the current period.
Based on the same application concept as the multi-target tracking method based on the automobile radar, the application provides a computer device, as described in the following embodiments. Because the principle of solving the problems of the computer equipment is similar to the multi-target tracking method based on the automobile radar, the implementation of the computer equipment can refer to the implementation of the multi-target tracking method based on the automobile radar, and repeated parts are not repeated.
A computer device including a memory, a processor and a computer program stored in the memory and executable on the processor, as shown in fig. 1, wherein the processor executes the computer program to implement the steps of the multi-target tracking method based on the automotive radar, and the steps include:
s101: and acquiring a plurality of detection targets corresponding to each vehicle to be detected in the current period, and clustering each detection target by using a density clustering algorithm to generate each effective target. Wherein the valid targets include: a valid target state.
S102: and calculating the correlation degree of each effective target and each acquired previous period flight path respectively to generate a first covariance matrix.
S103: and generating an evaluation matrix by utilizing dynamic alpha filtering according to the first covariance matrix and the acquired second covariance matrix in the previous period.
S104: and performing Hungarian assignment according to the priority and the evaluation matrix of each previous period of flight path to generate a matched flight path corresponding to each effective target. Wherein, the track of matching includes: and matching the track state.
S105: and performing Kalman filtering according to each effective target state, each previous period flight path and the previous period second covariance matrix to generate a current period second covariance matrix.
S106: and generating the current period flight path of each effective target and the object set after resampling in the current period by utilizing Monte Carlo multiple probability sampling according to the state of each effective target, the matched flight path state corresponding to each effective target, the first covariance matrix and the second covariance matrix in the current period.
Based on the same application concept as the multi-target tracking method based on the automotive radar, the present application provides a computer-readable storage medium, as described in the following embodiments. Because the principle of solving the problems of the computer-readable storage medium is similar to the multi-target tracking method based on the automobile radar, the implementation of the computer-readable storage medium can refer to the implementation of the multi-target tracking method based on the automobile radar, and repeated parts are not described again.
A computer-readable storage medium, on which a computer program is stored, as shown in fig. 1, the computer program, when executed by a processor, implementing the steps of the above-mentioned multi-target tracking method based on automotive radars, comprising:
s101: and acquiring a plurality of detection targets corresponding to each vehicle to be detected in the current period, and clustering each detection target by using a density clustering algorithm to generate each effective target. Wherein the valid targets include: a valid target state.
S102: and calculating the correlation degree of each effective target and each acquired previous period flight path respectively to generate a first covariance matrix.
S103: and generating an evaluation matrix by utilizing dynamic alpha filtering according to the first covariance matrix and the acquired second covariance matrix in the previous period.
S104: and performing Hungarian assignment according to the priority and the evaluation matrix of each previous period of flight path to generate a matched flight path corresponding to each effective target. Wherein, the matching track includes: and matching the track state.
S105: and performing Kalman filtering according to each effective target state, each previous period flight path and the previous period second covariance matrix to generate the current period second covariance matrix.
S106: and generating the current period flight path of each effective target and the object set after resampling in the current period by utilizing Monte Carlo multiple probability sampling according to the state of each effective target, the matched flight path state corresponding to each effective target, the first covariance matrix and the second covariance matrix in the current period.
As will be appreciated by one skilled in the art, embodiments of the present invention may be provided as a method, system, or computer program product. Accordingly, the present invention may take the form of an entirely hardware embodiment, an entirely software embodiment or an embodiment combining software and hardware aspects. Furthermore, the present invention may take the form of a computer program product embodied on one or more computer-usable storage media (including, but not limited to, disk storage, CD-ROM, optical storage, and the like) having computer-usable program code embodied therein.
The present invention is described with reference to flowchart illustrations and/or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of the invention. It will be understood that each flow and/or block of the flow diagrams and/or block diagrams, and combinations of flows and/or blocks in the flow diagrams and/or block diagrams, can be implemented by computer program instructions. These computer program instructions may be provided to a processor of a general purpose computer, special purpose computer, embedded processor, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, create means for implementing the functions specified in the flowchart flow or flows and/or block diagram block or blocks.
These computer program instructions may also be stored in a computer-readable memory that can direct a computer or other programmable data processing apparatus to function in a particular manner, such that the instructions stored in the computer-readable memory produce an article of manufacture including instruction means which implement the function specified in the flowchart flow or flows and/or block diagram block or blocks.
These computer program instructions may also be loaded onto a computer or other programmable data processing apparatus to cause a series of operational steps to be performed on the computer or other programmable apparatus to produce a computer implemented process such that the instructions which execute on the computer or other programmable apparatus provide steps for implementing the functions specified in the flowchart flow or flows and/or block diagram block or blocks.
The principle and the implementation mode of the invention are explained by applying specific embodiments in the invention, and the description of the embodiments is only used for helping to understand the method and the core idea of the invention; meanwhile, for a person skilled in the art, according to the idea of the present invention, there may be variations in the specific embodiments and the application scope, and in summary, the content of the present specification should not be construed as a limitation to the present invention.
Claims (9)
1. A multi-target tracking method based on an automobile radar is characterized by comprising the following steps:
acquiring a plurality of detection targets corresponding to each vehicle to be detected in the current period, and clustering each detection target by using a density clustering algorithm to generate each effective target; the effective targets include: a valid target state;
calculating the correlation degree of each effective target and each acquired previous period flight path respectively to generate a first covariance matrix;
generating an evaluation matrix by utilizing dynamic alpha filtering according to the first covariance matrix and the acquired second covariance matrix in the previous period;
performing Hungary assignment according to the priority of each previous period flight path and the evaluation matrix to generate a matched flight path corresponding to each effective target; the matching track comprises: matching a track state;
performing Kalman filtering according to each effective target state, each previous period flight path and the previous period second covariance matrix to generate a current period second covariance matrix;
and generating a current period flight path of each effective target and a current period resampled object set by utilizing Monte Carlo multiple probability sampling according to the state of each effective target, the corresponding matched flight path state of each effective target, the first covariance matrix and the current period second covariance matrix.
2. The automotive radar-based multi-target tracking method according to claim 1, wherein each previous cycle track comprises: track cycle number, accumulated error value and previous cycle track state.
3. The automotive radar-based multi-target tracking method according to claim 2, further comprising:
and calculating the corresponding priority of each previous period track according to the track cycle number of each previous period track and the accumulated error value.
4. The multi-target tracking method based on automotive radars of claim 3, wherein the priority
The calculation formula is specifically as follows:
wherein i is the ith track in the previous cycle track, k is the current cycle value, the value range of k is a positive integer greater than or equal to 1,
the number of track cycles of the ith track of the previous track,
and accumulating error values for the flight path of the ith flight path of the previous period of flight path.
5. The automotive radar-based multi-target tracking method according to claim 1, wherein the generating of the current period track and the current period resampled object set of each effective target by using monte carlo multivariate probability sampling according to each effective target state, the matched track state corresponding to each effective target, the first covariance matrix and the current period second covariance matrix comprises:
performing weighted calculation according to the state of each effective target, the matched track state corresponding to each effective target and the first covariance matrix to generate a sampling center;
sampling by using a sampling function according to the obtained object set subjected to resampling in the previous period, the sampling center, the first covariance matrix and the second covariance matrix in the current period to generate a sampling object; the sampling object includes: object state and object covariance;
performing weight calculation according to the object state and the object covariance to generate expectation of each sampling object, and updating the current periodic track of each effective target according to the expectation of each sampling object;
and resampling by utilizing a preset sampling rule according to the sampling center to generate an object set resampled in the current period.
6. The automotive radar-based multi-target tracking method according to claim 1, wherein the valid target states include: distance, velocity, angle, and acceleration.
7. A multi-target tracking system based on automotive radar is characterized by comprising:
the clustering unit is used for acquiring a plurality of detection targets corresponding to each vehicle to be detected in the current period, and clustering each detection target by using a density clustering algorithm to generate each effective target; the effective targets include: a valid target state;
the correlation unit is used for calculating the correlation between each effective target and each acquired previous period flight path respectively to generate a first covariance matrix;
the dynamic filtering unit is used for generating an evaluation matrix by utilizing dynamic alpha filtering according to the first covariance matrix and the acquired second covariance matrix of the previous period;
the assignment unit is used for performing Hungarian assignment according to the priority of each previous period flight path and the evaluation matrix to generate a matched flight path corresponding to each effective target; the matching track comprises: matching a track state;
the Kalman filtering unit is used for carrying out Kalman filtering according to each effective target state, each previous period flight path and the previous period second covariance matrix to generate a current period second covariance matrix;
and the sampling unit is used for generating the current period flight path of each effective target and the object set after the current period resampling by utilizing Monte Carlo multiple probability sampling according to the state of each effective target, the corresponding matched flight path state of each effective target, the first covariance matrix and the current period second covariance matrix.
8. A computer device comprising a memory, a processor and a computer program stored on the memory and executable on the processor, characterized in that the processor implements the steps of the automotive radar-based multi-target tracking method according to any one of claims 1 to 6 when executing the program.
9. A computer-readable storage medium, on which a computer program is stored, which, when being executed by a processor, carries out the steps of the automotive radar-based multi-target tracking method according to any one of claims 1 to 6.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201811423573.XA CN109581353B (en) | 2018-11-27 | 2018-11-27 | Multi-target tracking method and system based on automobile radar |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201811423573.XA CN109581353B (en) | 2018-11-27 | 2018-11-27 | Multi-target tracking method and system based on automobile radar |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN109581353A CN109581353A (en) | 2019-04-05 |
| CN109581353B true CN109581353B (en) | 2022-11-25 |
Family
ID=65924776
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201811423573.XA Active CN109581353B (en) | 2018-11-27 | 2018-11-27 | Multi-target tracking method and system based on automobile radar |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN109581353B (en) |
Families Citing this family (9)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN110361744B (en) * | 2019-07-09 | 2022-11-01 | 哈尔滨工程大学 | RBMCDA underwater multi-target tracking method based on density clustering |
| CN110780289B (en) * | 2019-10-23 | 2021-07-30 | 北京信息科技大学 | Multi-target vehicle tracking method and device based on scene radar |
| CN111551938B (en) * | 2020-04-26 | 2022-08-30 | 北京踏歌智行科技有限公司 | Unmanned technology perception fusion method based on mining area environment |
| CN112590808B (en) * | 2020-12-23 | 2022-05-17 | 东软睿驰汽车技术（沈阳）有限公司 | Multi-sensor fusion method and system and automatic driving vehicle |
| CN113093139B (en) * | 2021-04-07 | 2022-05-06 | 烟台大学 | Data association method for multi-target tracking |
| CN113391270B (en) * | 2021-06-11 | 2022-10-04 | 森思泰克河北科技有限公司 | False target suppression method and device for multi-radar point cloud fusion and terminal equipment |
| CN113514810B (en) * | 2021-07-07 | 2023-07-18 | 北京信息科技大学 | MIMO radar observation noise optimization method and device |
| CN114019497B (en) * | 2022-01-05 | 2022-03-18 | 南京楚航科技有限公司 | Target lane change identification method based on millimeter wave radar variance statistics |
| CN114859339B (en) * | 2022-04-29 | 2023-06-02 | 电子科技大学 | Multi-target tracking method based on millimeter wave radar |
Family Cites Families (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| JP2893176B2 (en) * | 1997-05-08 | 1999-05-17 | 防衛庁技術研究本部長 | Multi-target tracking device |
| CN101770024B (en) * | 2010-01-25 | 2011-08-31 | 上海交通大学 | Multi-target tracking method |
| CN105137418B (en) * | 2015-07-28 | 2017-06-16 | 中国人民解放军海军航空工程学院 | Multiple target tracking and data interconnection method based on complete adjacent fuzzy clustering |
| CN106372646B (en) * | 2016-08-30 | 2020-07-14 | 上海交通大学 | Multi-target tracking method based on SRCK-GMCPHD filtering |
| CN106772351B (en) * | 2016-11-16 | 2019-04-23 | 西安电子科技大学 | Kalman filter method based on the memory of limited step |
-
2018
- 2018-11-27 CN CN201811423573.XA patent/CN109581353B/en active Active
Also Published As
| Publication number | Publication date |
|---|---|
| CN109581353A (en) | 2019-04-05 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN109581353B (en) | Multi-target tracking method and system based on automobile radar | |
| CN110596694B (en) | Complex environment radar multi-target tracking and road driving environment prediction method | |
| Hammarstrand et al. | Extended object tracking using a radar resolution model | |
| Daun et al. | Tracking in multistatic passive radar systems using DAB/DVB-T illumination | |
| CN110780289B (en) | Multi-target vehicle tracking method and device based on scene radar | |
| Knill et al. | A direct scattering model for tracking vehicles with high-resolution radars | |
| CN109508000A (en) | Isomery multi-sensor multi-target tracking method | |
| US20100109938A1 (en) | Adaptive radar | |
| Eltrass et al. | Automotive radar system for multiple‐vehicle detection and tracking in urban environments | |
| Gunnarsson et al. | Tracking vehicles using radar detections | |
| CN113536850B (en) | Target object size testing method and device based on 77G millimeter wave radar | |
| CN113408504B (en) | Lane line identification method and device based on radar, electronic equipment and storage medium | |
| Holder et al. | Modeling and simulation of radar sensor artifacts for virtual testing of autonomous driving | |
| Polychronopoulos et al. | Integrated object and road border tracking using 77 GHz automotive radars | |
| CN111007880B (en) | Extended target tracking method based on automobile radar | |
| Lamard et al. | Multi target tracking with cphd filter based on asynchronous sensors | |
| Cook et al. | Real-time radar-based tracking and state estimation of multiple non-conformant aircraft | |
| Kumar et al. | A novel algorithm to track closely spaced road vehicles using a low density flash lidar | |
| CN116699596A (en) | Method and device for correcting speed of millimeter wave radar of vehicle | |
| Schuster et al. | Tracking of vehicles on nearside lanes using multiple radar sensors | |
| CN115220002A (en) | Multi-target data association tracking method and related device for fixed single station | |
| JP2020008351A (en) | Tracking device, method for tracking, and program | |
| Huang et al. | Convoy tracking in Doppler blind zone regions using GMTI radar | |
| Metzner et al. | On multi-sensor radar configurations for vehicle tracking in autonomous driving environments | |
| Pollard et al. | Convoy detection processing by using the hybrid algorithm (gmcphd/vs-immc-mht) and dynamic bayesian networks |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| GR01 | Patent grant | ||
| GR01 | Patent grant |