A vehicle-road cooperation multi-target tracking method and system

Abstract

The application relates to a vehicle-road cooperation multi-target tracking method and system, and the method comprises the following steps: step S1, a host vehicle and a road end respectively acquire self position data, and acquire sensing data of surrounding target vehicles through sensors; step S2, based on the sensing data of the target vehicles, an adaptive Gaussian mixture probability density hypothesis algorithm is used to respectively solve tracking results of the target vehicles sensed by the host vehicle and the road end in a global coordinate system; step S3, based on position information and a time stamp, time and space synchronization is performed on the tracking results of the target vehicles; step S4, a data association algorithm is used to perform data association on the tracking results of the target vehicles sensed by the host vehicle and the road end after time and space synchronization, a multi-source fusion algorithm is used for fusion, and a tracking result after vehicle-road cooperation fusion is output. Compared with the prior art, the application solves the problem that the number of vehicle targets in an actual traffic scene is unknown and randomly appears, and has the advantages of small calculation amount and high multi-target tracking precision.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of intelligent transportation, in particular to a vehicle-road cooperation multi-target tracking method and system. BACKGROUND

[0002] At present, the single vehicle environment perception system mainly uses vehicle-mounted sensors to obtain the surrounding environment of the vehicle, such as laser radar, millimeter wave radar and camera, etc. The vehicle-mounted computing unit can improve the accuracy of target state estimation by calculating the fusion results of multiple sensors, but the working principles of different sensors are different, the anti-interference abilities are different, and the time base and sampling period of each sensor are also quite different, so there are great challenges in data fusion between different sensors, especially in the fusion of raw data and the fusion of feature layer data. In addition, the sensor is limited by the physical characteristics, and the perception range is limited, and it is also restricted by the installation position. In a complex road traffic environment, the vehicle-mounted sensor is easily affected by the mutual shielding of targets and the shielding of obstacles, and cannot eliminate the driving blind area, and the perception range is limited.

[0003] Through retrieval, Chinese patent application CN113486300A discloses a multi-target tracking method for unmanned vehicles. The method first perceives the surrounding environment through the self-vehicle sensor, estimates the state of a single target using the lossless Kalman filtering algorithm, and then correlates the target state using the joint probability data association algorithm to realize multi-target tracking. However, this method cannot adaptively estimate new appearing targets, and thus cannot solve the problem of unknown and random appearing of vehicle targets in actual traffic scenarios.

[0004] Chinese patent application CN112946624A discloses a multi-target tracking algorithm based on a track management method. The method uses a multi-hypothesis tracking algorithm to realize track initiation, track association, track merging and track deletion. This multi-target tracking algorithm based on data association will rapidly increase the calculation amount with the increase of the number of targets and clutter, and even cause the problem of combination explosion.

[0005] In addition, the above-mentioned methods are all multi-target tracking algorithms based on single vehicle intelligence, which are easily affected by the mutual shielding of targets and the shielding of obstacles, cannot eliminate the driving blind area, and have limited perception range. SUMMARY

[0006] The present application provides a vehicle-road cooperation multi-target tracking method and system to overcome the defects of the prior art.

[0007] The object of the present application can be achieved by the following technical solutions:

[0008] According to a first aspect of the present application, a vehicle-road cooperation multi-target tracking method is provided, which comprises the following steps:

[0009] Step S1, the host vehicle and the road end obtain their own position data respectively, and obtain the perception data of the target vehicle in the surrounding through sensors respectively;

[0010] Step S2, based on the perception data of the target vehicle, the adaptive Gaussian mixture probability density hypothesis algorithm is used to solve the tracking results of the target vehicle perceived by the host vehicle and the road end in the global coordinate system respectively;

[0011] Step S3, based on the position information and the time stamp, the tracking results of the target vehicle are time-space synchronized;

[0012] Step S4, the data association algorithm is used to associate the tracking results of the target vehicle perceived by the host vehicle and the road end after time-space synchronization, and the multi-source fusion algorithm is used for fusion, and the tracking results after vehicle-road cooperative fusion are output.

[0013] Preferably, the host vehicle and the road end in the step S1 obtain their own position data respectively, specifically: obtaining the position data of the host vehicle from the vehicle positioning system obtaining the position data of the road end from the road side unit wherein, respectively represent the horizontal position and the vertical position of the host vehicle relative to the global coordinate system in the Cartesian coordinate system, is the rotation angle of the host vehicle relative to the global coordinate system, respectively represent the horizontal position and the vertical position of the road end relative to the global coordinate system in the Cartesian coordinate system, is the rotation angle of the road end relative to the global coordinate system.

[0014] Preferably, the host vehicle and the road end in the step S1 obtain the perception data of the target vehicle in the surrounding through sensors, specifically:

[0015] The host vehicle and the road end obtain the perception data of the target vehicle in the surrounding through sensors respectively respectively represent the number of target vehicles perceived by the host vehicle and the road end at time k, respectively represent the perception data of the kth target vehicle by the host vehicle and the perception data of the kth target vehicle by the road end;

[0016] wherein, the perception data of the target vehicle includes the position data of the target vehicle perceived in the local coordinate system of the host vehicle and the road end.

[0017] Preferably, the step S2 includes the following sub-steps:

[0018] Step S21, define that the number of clutter or false detections obeys the Poisson distribution, and realize the adaptive birth of the target by using the Poisson point process model, specifically:​​

[0019] Posterior intensity v of target at time k-1 k-1|k-1 (x):

[0020]

[0021] where J k-1|k-1 is the number of Gaussian components at time k-1, is the weight corresponding to the i-th Gaussian component at time k-1, is the mean and covariance matrix of the i-th Gaussian component at time k-1, respectively;

[0022] The state transition probability density and the observation likelihood function are defined as Gaussian distributions:

[0023] f k|k-1 (x|ζ)=N(x;F k-1 ζ,Q k-1 )

[0024] g k (z|x)=N(z;H k x,R k )

[0025] where ζ is the state of the target, z is the observation of the target; F k-1 and H k are the state transition matrix and the observation matrix, respectively; Q k-1 and R k are the covariance matrices of the process noise and the observation noise, respectively;

[0026] In step S22, a recursive form of Gaussian mixture probability density hypothesis is obtained by using the probability density hypothesis algorithm, and tracking results of the target vehicles sensed by the host vehicle and the road-side sensor in the global coordinate system are solved, respectively.

[0027] Preferably, the step S22 comprises the following sub-steps:

[0028] In step S221, a prediction step is performed.

[0029] Predicted intensity v k|k-1 (x) of target:

[0030] v k|k-1 (x)=v S，k|k-1 (x)+γ k (x)

[0031] where γ k (x) is the intensity of the new-born target; v S，k|k-1 (x) is the predicted intensity of the surviving target;

[0032] Predicted intensity vS，k|k-1 The expression of (x) is:

[0033]

[0034] In the formula, P S，k is the existence probability of the target state at time k, is the weight of the i-th target at time k-1; represents that the target state prediction is subject to a Gaussian distribution with a mean of and a covariance of and satisfies:

[0035]

[0036]

[0037] In the formula, F k-1 is the state transition matrix at time k-1, Q k-1 is the covariance matrix of the process noise at time k-1; is the mean of the i-th target state at time k-1, is the covariance of the i-th target state at time k-1;

[0038] The intensity of the new target is γ k The expression of (x) is:

[0039]

[0040] In the formula, M k is the number of target vehicles perceived by the host vehicle or the road end at time k, w (i) is the initial weight of the i-th target vehicle; represents that the target state is subject to a Gaussian distribution with a mean of and a covariance of P0, wherein P0 is an initial covariance matrix, is the state of the target vehicle in the global coordinate system constructed adaptively according to the perception data z k，i of the detected target vehicle, and satisfies:

[0041]

[0042] In the formula, represents the rotation matrix of the host vehicle or the road end relative to the global coordinate system at time k, θ k is the rotation angle of the host vehicle or the road end relative to the global coordinate system at time k; represents the translation matrix of the host vehicle or the road end relative to the global coordinate system at time k, wherein ξ k and η k respectively represent the horizontal position and the vertical position of the host vehicle or the road end relative to the global coordinate system in the Cartesian coordinate system.

[0043] Step S222, Update Step:

[0044] The update strength v of the target at time k-1 k|k (x):

[0045]

[0046]

[0047]

[0048]

[0049] In the formula, v k|k-1 (x) represents the predicted intensity of the target, P D，k Let k be the probability of detection at time k. The weights are updated for the j-th target state, and z is the observed value of the target. Let H be the mean and covariance of the updated target state for the j-th objective, respectively; k ′ is the observation matrix after coordinate correction, z G Let I represent the observed state of the target in the global coordinate system, where I is the identity matrix and K is the Kalman gain, satisfying:

[0050] H k′ =R(θ) k )H k

[0051] z G =R(θ) k )z L +T k

[0052]

[0053] In the formula, θ represents the rotation matrix of the main vehicle or road end relative to the global coordinate system. k H is the relative rotation angle between the main vehicle and the road end. k The observation matrix before correction; Let ξ represent the translation matrix of the main vehicle or road end relative to the global coordinate system, where ξ is the translation matrix of the main vehicle or road end relative to the global coordinate system. k η k These represent the horizontal and vertical positions of the main vehicle or road end relative to the global coordinate system in the Cartesian coordinate system, respectively; z L This represents the observation data of the main vehicle or the road end in their respective local coordinate systems;

[0054] Step S223: Update the target state set according to the adaptive Gaussian mixture probability density hypothesis algorithm. The pruning and merging algorithm is used to filter the state of targets, mainly involving two processes: deletion and merging.

[0055] 1) Based on the deletion threshold value D th Remove weights Gaussian terms, retaining weights Gaussian terms;

[0056] 2) Based on the merging threshold value M th Calculate the distance of the Gaussian term. d ij <M th Merge the Gaussian terms;

[0057] After updating the weights, states, and error covariance matrices of the corresponding items in steps S221-S222, the multi-objective state set after deletion and merging is obtained. and in, Tracking results of targets sensed by the main vehicle and the roadside in the global coordinate system. and These represent the number of target vehicles tracked by the main vehicle and the roadside vehicle, respectively.

[0058] Preferably, step S4 includes the following sub-steps:

[0059] Step S41: Use the Hungarian data association algorithm to associate the tracking results from the main vehicle and the roadside;

[0060] Step S42: Use the fast covariance cross-fusion algorithm to fuse the successfully associated tracking results and output the tracking results after vehicle-road cooperative fusion.

[0061] Preferably, the Hungarian data association algorithm used in step S41 to associate the tracking results from the main vehicle and the roadside specifically involves:

[0062] 1) The problem of associating the master vehicle with the roadside tracking target is transformed into the following allocation optimization problem:

[0063]

[0064]

[0065]

[0066] in, The target being tracked by the main vehicle and the target of roadside tracking Euclidean distance between them; u ij The target being tracked by the main vehicle and the target of roadside tracking The binary decision parameters between them and These represent the number of target vehicles tracked by the main vehicle and the roadside vehicle, respectively.

[0067] 2) Solve the allocation optimization problem using the Hungarian data association algorithm to obtain the correspondence between the master vehicle and the target vehicle being tracked on the roadside, and then associate the tracking results.

[0068] Preferably, the successful tracking results fused using the fast covariance cross-fusion algorithm in step S42 are specifically as follows:

[0069]

[0070]

[0071]

[0072]

[0073] In the formula, P1 and P2 are two state estimates of the same target from the main vehicle and the roadside, respectively. P1 and P2 are the covariances of the state estimates of the main vehicle and the roadside for the same target, respectively. l represents the dimension of the state. tr is the trace of the matrix.

[0074] Store the target state where the master vehicle and the roadside are successfully associated and merged. N k This indicates the number of targets after fusion, and stores the states of targets that were not successfully associated. Finally, the tracking results after vehicle-road cooperative fusion are output.

[0075] According to a second aspect of the present invention, a vehicle-road cooperative multi-target tracking system is provided, employing any of the methods described above. The system includes a vehicle-end perception module, a time-space synchronization module, and a data association and data fusion module located at the vehicle end, and a road-end perception module located at the road end; the vehicle end and the road end communicate with each other through a vehicle-road communication module.

[0076] The vehicle-side perception module includes an onboard sensor perception module (11) for acquiring target perception data around the main vehicle, an onboard positioning module for acquiring main vehicle location data, a first target detection module for acquiring the state of target objects around the main vehicle using pattern recognition technology, and a first target tracking module for tracking the state of the detected target objects.

[0077] The roadside perception module includes a roadside sensor perception module for acquiring perception data of targets around the roadside, a second target detection module for acquiring the state of targets around the roadside using pattern recognition technology, and a second target tracking module for tracking the state of the detected targets.

[0078] The vehicle-to-infrastructure communication module includes a roadside communication unit located at the road end and an in-vehicle communication unit located at the vehicle end, used to send the roadside tracking results, location information and timestamp information obtained at the road end to the in-vehicle unit;

[0079] The time-space synchronization module is used to synchronize the target vehicle tracking results obtained from the vehicle end and the road end in time and space.

[0080] Preferably, the roadside communication unit uses C-V2X technology to communicate with the vehicle-mounted communication unit.

[0081] Compared with the prior art, the present invention has the following advantages:

[0082] 1) The method of the present invention uses an adaptive Gaussian mixture probability density hypothesis algorithm to solve the tracking results of the target vehicle perceived by the main vehicle and the roadside in the global coordinate system, and uses a Poisson point process model to realize the adaptive birth of the target, which can solve the problem that the number of vehicle targets is unknown and appears randomly in actual traffic scenarios.

[0083] 2) By pruning and fusing the updated state and weights using the adaptive Gaussian mixture probability density hypothesis algorithm, the computational load will not increase rapidly as the number of targets and clutter increases, thus avoiding the situation of combinatorial explosion and further improving the accuracy of multi-target tracking.

[0084] 3) The adoption of a vehicle-road cooperative multi-target tracking algorithm not only improves the accuracy of multi-target tracking, but also reduces the driving blind spot and expands the vehicle's perception range. Attached Figure Description

[0085] Figure 1 This is a flowchart of the method of the present invention;

[0086] Figure 2 This is a schematic diagram of the system modules of the present invention;

[0087] Figure 3 This is a schematic diagram of a scenario in the embodiment;

[0088] Reference numerals: 1-Vehicle-side perception module, 11-Vehicle-mounted sensor perception module, 12-Vehicle-mounted positioning module, 13-First target detection module, 14-First target tracking module; 2-Roadside perception module, 21-Roadside sensor perception module, 22-Second target detection module, 23-Second target tracking module; 3-Vehicle-road communication module; 4-Time-space synchronization module, 5-Data association module, 6-Data fusion module. Detailed Implementation

[0089] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some, not all, of the embodiments of the present invention. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort should fall within the scope of protection of the present invention.

[0090] Example

[0091] First, an embodiment of the method of the present invention is given, namely a vehicle-road cooperative multi-target tracking method, such as... Figure 1 As shown, the method includes the following steps:

[0092] Step S1: The main vehicle and the roadside vehicle acquire their own location data, and obtain perception data of surrounding target vehicles through sensors, specifically as follows:

[0093] Obtain the vehicle's location data from the vehicle positioning system. Obtain location data of the roadside from the positioning system. in, These represent the horizontal and vertical positions of the main vehicle relative to the global coordinate system in the Cartesian coordinate system, respectively. The rotation angle of the master vehicle relative to the global coordinate system. These represent the horizontal and vertical positions of the lower end of the Cartesian coordinate system relative to the global coordinate system, respectively. The rotation angle of the road end relative to the global coordinate system.

[0094] Both the main vehicle and the roadside vehicle acquire perception data of surrounding target vehicles through sensors. These represent the number of target vehicles detected by the main vehicle and the roadside at time k, respectively. They represent the relationship between the master vehicle and the first vehicle at time k. The perception data of the target vehicle, the roadside connection to the first The perception data of the target vehicle includes the target vehicle's position data perceived in the local coordinate system of the main vehicle and the roadside.

[0095] Step S21: Define the number of clutter or false detections as following a Poisson distribution, and use a Poisson point process model to achieve adaptive target birth, specifically:

[0096] The posterior strength v of the target at time k-1 k-1|k-1 (x):

[0097]

[0098] In the formula, J k-1|k-1 This represents the number of Gaussian components at time k-1. This represents the weight of the i-th Gaussian component at time k-1. Let be the mean and covariance matrix of the i-th Gaussian component at time k-1, respectively.

[0099] Define the state transition probability density and observation likelihood function as following a Gaussian distribution:

[0100] f k|k-1 (x|ζ)=N(x;F k-1 ζ, Q k-1 )

[0101] g k (z|x)=N(z;H k x, R k )

[0102] In the formula, ζ represents the target's state, and z represents the target's observed value; F k-1 and H k These are the state transition matrix and the observation matrix, Q. k-1 and R k These are the covariance matrices of the process noise and the observation noise, respectively.

[0103] Step S22: Using the probability density hypothesis algorithm, obtain the recursive form of the Gaussian mixture probability density hypothesis, and solve for the tracking results of the target vehicle sensed by the main vehicle and the roadside in the global coordinate system, including the following sub-steps:

[0104] Step S221, Prediction Step:

[0105] Target prediction intensity v k|k-1 (x):

[0106] v k|k-1 (x)=v S，k|k-1 (x)+γ k (x)

[0107] In the formula, γ k (x) represents the intensity of the new target; v S，k|k-1 (x) represents the predicted strength of the surviving target;

[0108] Predicted intensity v of surviving targets S，k|k-1 The expression for (x) is:

[0109]

[0110] In the formula, P S，k It is the probability of the existence of the target state at time k. Let be the weight of the i-th target at time k-1; The target state prediction follows the mean. covariance is It has a Gaussian distribution and satisfies:

[0111]

[0112]

[0113] In the formula, F k-1 Let Q be the state transition matrix at time k-1. k-1 Let be the covariance matrix of the process noise at time k-1; Let be the mean of the i-th target state at time k-1. Let be the covariance of the i-th target state at time k-1;

[0114] Intensity γ of newly formed targets k The expression for (x) is:

[0115]

[0116] In the formula, M k Let w be the number of target vehicles sensed by the main vehicle or the roadside at time k. (i) Let be the initial weight of the i-th target vehicle; The representative obeys the mean. The target state is a Gaussian distribution with covariance P0, where P0 is the initial covariance matrix. Based on the perception data z of the detected target vehicle k，i The adaptively constructed state of the target vehicle in the global coordinate system satisfies:

[0117]

[0118] In the formula, Let θ represent the rotation matrix of the main vehicle or road end relative to the global coordinate system at time k. k Let k be the relative rotation angle between the main vehicle and the road end. Let ξ represent the translation matrix of the main vehicle or road end relative to the global coordinate system at time k, where ξ k η kThese represent the horizontal and vertical positions of the main vehicle or road end relative to the global coordinate system in the Cartesian coordinate system, respectively.

[0119] Step S222, Update Step:

[0120] The update strength v of the target at time k-1 k|k (x):

[0121]

[0122]

[0123]

[0124]

[0125] In the formula, v k|k-1 (x) represents the predicted intensity of the target, P D，k P is the probability of detection at time k, in this embodiment. D，k =0.98; The weights are updated for the j-th target state, and z is the observed value of the target. Let H be the mean and covariance of the updated target state for the j-th objective, respectively; k ′ is the observation matrix after coordinate correction, z G Let I represent the observed state of the target in the global coordinate system, where I is the identity matrix and K is the Kalman gain, satisfying:

[0126] H k ′=R(θ k )H k

[0127] z G =R(θ) k )z L +T k

[0128]

[0129] In the formula, θ represents the rotation matrix of the main vehicle or road end relative to the global coordinate system. k H is the rotation angle of the main vehicle or road end relative to the global coordinate system. k The observation matrix before correction; Let ξ represent the translation matrix of the main vehicle or road end relative to the global coordinate system, where ξ is the translation matrix of the main vehicle or road end relative to the global coordinate system. k η k These represent the horizontal and vertical positions of the main vehicle or road end relative to the global coordinate system in the Cartesian coordinate system, respectively; z L This represents the observation data of the main vehicle or the road end in their respective local coordinate systems;

[0130] Step S223: Update the target state set according to the adaptive Gaussian mixture probability density hypothesis algorithm. The pruning and merging algorithm is used to filter the state of targets, mainly involving two processes: deletion and merging.

[0131] 1) Based on the deletion threshold value D th Remove weights Gaussian terms, retaining weights Gaussian terms, in this embodiment D th =5×10 -5 ;

[0132] 2) Based on the merging threshold value M th Calculate the distance of the Gaussian term. d ij <M th The Gaussian terms are merged in this embodiment, M. th =2;

[0133] After updating the weights, states, and error covariance matrices of the corresponding items in steps S221-S222, the multi-objective state set after deletion and merging is obtained. and in, Tracking results of targets sensed by the main vehicle and the roadside in the global coordinate system. and These represent the number of target vehicles tracked by the main vehicle and the roadside vehicle, respectively.

[0134] Step S3: Based on location information and timestamps, synchronize the tracking results of the target vehicle in time and space;

[0135] Step S4: Use a data association algorithm to associate the tracking results of the master vehicle and the target vehicle sensed by the roadside after time and space synchronization, and use a data fusion algorithm to fuse them and output the tracking results after vehicle-road cooperative fusion.

[0136] Step S41: Use the Hungarian data association algorithm to associate the tracking results from the main vehicle and the roadside;

[0137] 1) The problem of associating the master vehicle with the roadside tracking target is transformed into the following allocation optimization problem:

[0138]

[0139]

[0140]

[0141] in, The target being tracked by the main vehicle and the target of roadside tracking Euclidean distance between them; u ij The target being tracked by the main vehicle and the target of roadside tracking The binary decision parameters between them and These represent the number of target vehicles tracked by the main vehicle and the roadside vehicle, respectively.

[0142] 2) Solve the allocation optimization problem using the Hungarian data association algorithm to obtain the correspondence between the master vehicle and the target vehicle being tracked on the roadside, and then associate the tracking results.

[0143] Step S42: Use the fast covariance cross-fusion algorithm to fuse the successfully associated tracking results and output the tracking results after vehicle-road cooperative fusion:

[0144]

[0145]

[0146]

[0147]

[0148] In the formula, P1 and P2 are two state estimates of the same target from the main vehicle and the roadside, respectively. P1 and P2 are the covariances of the state estimates of the main vehicle and the roadside for the same target, respectively. l represents the dimension of the state. tr is the trace of the matrix.

[0149] Store the target state where the master vehicle and the roadside are successfully associated and merged. N k This indicates the number of targets after fusion, and stores the states of targets that were not successfully associated. Finally, the tracking results after vehicle-road cooperative fusion are output.

[0150] Next, combined Figure 2 and Figure 3 This invention provides a system embodiment of a vehicle-road cooperative multi-target tracking system, employing the aforementioned method, including a vehicle-end perception module 1, a time-space synchronization module 4, a data association module 5, and a data fusion module 6 placed on the vehicle end, and a roadside (e.g., Figure 3 The roadside sensing module 2 of the RSU (Roadside Unit) in the vehicle; the vehicle end and the roadside end communicate through the vehicle-road communication module 3.

[0151] The vehicle-side perception module 1 includes an onboard sensor perception module 11 for acquiring target perception data around the main vehicle, an onboard positioning module 12 for acquiring the main vehicle's position data, a first target detection module 13 for acquiring the state of target objects around the main vehicle using pattern recognition technology, and a first target tracking module 14 for tracking the state of the detected target objects.

[0152] The roadside perception module 2 includes a roadside sensor perception module 21 for acquiring perception data of targets around the roadside, a second target detection module 22 for acquiring the state of targets around the roadside using pattern recognition technology, and a second target tracking module 23 for tracking the state of the detected targets.

[0153] The vehicle-road communication module 3 includes a roadside communication unit 31 placed at the road end and an on-board communication unit 32 placed at the vehicle end, which is used to send the roadside tracking results, location information and timestamp information obtained at the road end to the on-board unit.

[0154] The time-space synchronization module 4 is used to synchronize the target vehicle tracking results obtained from the vehicle end and the road end in time and space.

[0155] The roadside communication unit 31 is connected to the vehicle-mounted communication unit 32 using C-V2X technology; C-V2X technology is a vehicle wireless communication technology evolved from cellular network communication technologies such as 3G / 4G / 5G.

[0156] The above description is merely a specific embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any person skilled in the art can easily conceive of various equivalent modifications or substitutions within the technical scope disclosed in the present invention, and these modifications or substitutions should all be covered within the scope of protection of the present invention. Therefore, the scope of protection of the present invention should be determined by the scope of the claims.

Claims

1. A vehicle-road cooperative multi-target tracking method, characterized in that, The method comprises the following steps: Step S1, the host vehicle and the road end obtain their own position data respectively, and obtain the sensing data of the target vehicle in the surrounding area through sensors respectively; The main vehicle and the roadside unit obtain perception data of surrounding target vehicles through sensors. Specifically, the main vehicle and the roadside unit respectively obtain perception data of surrounding target vehicles through sensors. , , , They are respectively The number of target vehicles detected by the main vehicle and the roadside at all times. , They represent The main car is at the first The perception data of the target vehicle, the roadside connection to the first The perception data of the target vehicle includes the position data of the target vehicle perceived in the coordinate system of the main vehicle and the local coordinate system of the road end. Step S2, based on the sensing data of the target vehicle, the adaptive Gaussian mixture probability density hypothesis algorithm is used to solve the tracking results of the target vehicle sensed by the host vehicle and the road end in the global coordinate system, comprising the following sub-steps: Step S21, define that the number of clutter or false detection obeys Poisson distribution, and realize adaptive birth of the target by using Poisson point process model, specifically: The goal is Posterior intensity at time : In the formula, represents the number of Gaussian components at the time point, represents the weight corresponding to the th Gaussian component at the time point, , , respectively, the mean and the covariance matrix of the th Gaussian component at the time point, . Define that the state transition probability density and the observation likelihood function obey Gaussian distribution: wherein the state of interest, the observation of interest; and are the state transition matrix and the observation matrix, respectively, and are the covariance matrices of the process noise and the observation noise, respectively. Step S22, use the probability density hypothesis algorithm to obtain the recursive form of the Gaussian mixture probability density hypothesis, and solve the tracking results of the target vehicle sensed by the host vehicle and the road end in the global coordinate system respectively; Step S3, based on the position information and the time stamp, the tracking results of the target vehicle are time and space synchronized; Step S4, the data association algorithm is used to associate the tracking results of the target vehicle sensed by the host vehicle and the road end after time and space synchronization, and the multi-source fusion algorithm is used for fusion, and the tracking results after vehicle-road cooperative fusion are output. 2.The vehicle-road cooperation multi-target tracking method according to claim 1, characterized in that, The host vehicle and the road end in the step S1 respectively acquire own position data, specifically: acquiring position data of the host vehicle from a vehicle positioning system , and acquiring position data of the road end from a road side unit ; wherein, , respectively represent horizontal position and vertical position of the host vehicle relative to a global coordinate system in a Cartesian coordinate system, is a rotation angle of the host vehicle relative to the global coordinate system, respectively represent horizontal position and vertical position of the road end relative to the global coordinate system in the Cartesian coordinate system, is a rotation angle of the road end relative to the global coordinate system. 3.The vehicle-road cooperation multi-target tracking method according to claim 1, characterized in that, The step S22 comprises the following sub-steps: Step S221, prediction step: Predicted intensity of target : wherein is the intensity of the new target; is the predicted intensity of the surviving target; Predicted intensity of a survival target The expression is: wherein is the probability of existence of the target state at time is the weight of the th target at time represents that the prediction of the target state obeys a Gaussian distribution with mean and covariance and satisfies: In the formula, for The state transition matrix at time t, for The covariance matrix of the time-series noise; for The first moment The mean of each target state, for Time of the first The covariance of each target state; intensity of the new target The expression is: wherein, is the number of target vehicles perceived by the host vehicle or the road side at the time instant, is the initial weight of the th target vehicle; represents the target state subject to a Gaussian distribution with mean and covariance , wherein is the initial covariance matrix, is the state of the target vehicle in the global coordinate system adaptively constructed according to the perception data of the detected target vehicle satisfies: wherein, denotes the rotation matrix of the host vehicle or the road side relative to the global coordinate system at time t, is the rotation angle of the host vehicle or the road side relative to the global coordinate system at time t; denotes the translation matrix of the host vehicle or the road side relative to the global coordinate system at time t, wherein, , denote the horizontal and vertical position of the host vehicle or the road side relative to the global coordinate system in the Cartesian coordinate system, respectively. Step S222, update step: The object is the update strength : wherein, the predicted intensity of the target, is the probability of detection at the time instant; is the weight of the th target state update, is the observation of the target, , is the mean and the covariance of the th target state update, respectively; is the observation matrix after coordinate correction, denotes the observation of the target state in the global coordinate system, is the identity matrix, is the Kalman gain, satisfying: wherein, represents a rotation matrix of the host vehicle or road end relative to the global coordinate system, is a relative rotation angle of the host vehicle or road end, is an observation matrix before correction; represents a translation matrix of the host vehicle or road end relative to the global coordinate system, wherein, , respectively represent a horizontal position and a vertical position of the host vehicle or road end relative to the global coordinate system under the Cartesian coordinate system; represents observation data of the host vehicle or road end under the respective local coordinate system; Step S223, updating the target state set according to the self-adaptive Gaussian mixture probability density hypothesis algorithm The pruning fusion algorithm is used to screen the state of the target, mainly including two processes of deletion and merging: 1) deleting the Gaussian terms with weights according to a deletion threshold , and keeping the Gaussian terms with weights ; 2) according to a merge threshold , calculate the distance of the Gaussians , merge the Gaussians of ; The weights, states and error covariance matrices of the corresponding items are updated through steps S221-S222 to obtain the multi-target state set after deletion and combination and wherein, 、 tracking results of the targets sensed by the host vehicle and the road end in the global coordinate system respectively, and respectively represent the number of target vehicles tracked by the host vehicle and the road end. 4.The vehicle-road cooperation multi-target tracking method according to claim 1, characterized in that, The step S4 comprises the following sub-steps: Step S41, the tracking results from the host vehicle and the road end are associated by using the Hungarian data association algorithm; Step S42, the fast covariance cross fusion algorithm is used to fuse the associated tracking results, and the tracking results after vehicle-road cooperative fusion are output.

5. The vehicle-road cooperation multi-target tracking method according to claim 4, characterized in that, In the step S41, the tracking results from the host vehicle and the road end are associated by using the Hungarian data association algorithm, specifically: 1) the association problem between the host vehicle and the road end tracking target is converted into the following assignment optimization problem: wherein, the target of the host vehicle tracking and the target of the road end tracking is the Euclidean distance between them; the target of the host vehicle tracking and the target of the road end tracking is the binary decision parameter between them, and respectively represent the number of target vehicles tracked by the host vehicle and the road end. 2) the Hungarian data association algorithm is used to solve the assignment optimization problem, and the corresponding relationship between the target vehicles tracked by the host vehicle and the road end is obtained, and the tracking results are associated.

6. The vehicle-road cooperation multi-target tracking method according to claim 4, characterized in that, In the step S42, the fast covariance cross fusion algorithm is used to fuse the associated tracking results, specifically: wherein are two state estimates of the same target by the host vehicle and the road side, respectively, are the corresponding covariances of the state estimates of the same target by the host vehicle and the road side, respectively; denotes the dimension of the state; is the trace of a matrix; Store the target state where the master vehicle and the roadside are successfully associated and merged. , This indicates the number of targets after fusion, and stores the states of targets that were not successfully associated. Finally, the tracking results after vehicle-road cooperative fusion are output. . 7.A vehicle infrastructure cooperative multi-target tracking system, characterized in that, The system comprises a vehicle-end sensing module (1), a time and space synchronization module (4), a data association module (5), a data fusion module (6) placed at the vehicle end, and a road-end sensing module (2) placed at the road end; the vehicle end and the road end communicate through the vehicle-road communication module (3) provided; The vehicle-end sensing module (1) comprises a vehicle-mounted sensor sensing module (11) for obtaining target sensing data around the host vehicle, a vehicle-mounted positioning module (12) for obtaining host vehicle position data, a first target detection module (13) for obtaining the state of the target around the host vehicle by using pattern recognition technology, and a first target tracking module (14) for tracking the state of the detected target; The road end perception module (2) comprises a road side sensor perception module (21) for acquiring target perception data around the road end, a second target detection module (22) for acquiring a state of a target around the road end by using a pattern recognition technique, and a second target tracking module (23) for tracking the state of the detected target; The vehicle-road communication module (3) comprises a road side communication unit (31) arranged at the road end and a vehicle-mounted communication unit (32) arranged at the vehicle end, and is used for transmitting the road side tracking result, position information and time stamp information obtained at the road end to the vehicle-mounted communication unit; The time and space synchronization module (4) is used for performing time and space synchronization on the target vehicle tracking results obtained at the vehicle end and the road end. 8.The vehicle infrastructure cooperative multi-target tracking system of claim 7, wherein, The road side communication unit (31) is in communication connection with the vehicle-mounted communication unit (32) by using a C-V2X technology.

Images