A graph matching based multi-sensor space-time calibration method for vehicle-road cooperative system

Abstract

The application relates to a kind of multi-sensor space-time calibration methods of vehicle-road cooperation system based on graph matching, comprising the following steps: S1, networking vehicle passes through the multi-sensor perception range of roadside, collects the self-positioning data of networking vehicle and its perception data to surrounding traffic flow, collects the perception data of multi-sensor to traffic flow;S2, using graph matching method, the vehicle ID collected by roadside multi-sensor is matched, and the trajectory of networking vehicle is identified in the multi-target trajectory data collected by roadside multi-sensor;S3, using Gauss-Newton method, the space-time parameters from the coordinate system of networking vehicle positioning data to the data coordinate system of roadside multi-sensor are solved;S4, according to the obtained space-time parameters, the collected data of roadside multi-sensor is transformed, and the space-time synchronization of roadside multi-sensor is realized.Compared with the prior art, the application has the advantages of fast, automatic parameter calibration of roadside multi-sensor, realization of space-time synchronization of roadside multi-sensor and the like.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of traffic information collection and analysis, and in particular to a multi-sensor space-time calibration method for a vehicle-road cooperation system based on graph matching. BACKGROUND

[0002] At present, intelligent construction of expressways, urban roads and the like is rapidly developing, and a large number of emerging sensors such as high-definition videos, millimeter wave radars and laser radars are deployed at road sides. However, the road side multi-sensors are asynchronous and heterogeneous in time and space dimensions, and in order to realize fine-grained fusion of road side sensing data, the position, posture and system time delay parameters of the road side sensors need to be accurately estimated, so that the road side multi-sensors are synchronized in time and space, and a good data basis is provided for further application.

[0003] Traditional multi-sensor space-time synchronization methods generally first realize clock synchronization of each sensor, and then calibrate the space parameters of each sensor. However, for a multi-sensor system deployed at a large range of road sides, the cost of a clock synchronization system is high. In a system without clock synchronization, the data collected by each road side sensor is not only in different space coordinate systems, but also has different system time delays. The traditional calibration method only calibrates the space parameters, and due to the existence of time delay, the calibration accuracy is reduced, and the road side multi-sensor data space-time synchronization effect is poor.

[0004] At the same time, the traditional method generally uses a customized calibration target, and through the movement of the target, the sensor can collect the trajectory data of the target, and the space-time parameters are calibrated by aligning the trajectory with the trajectory data obtained by the positioning device carried by the target. This method needs to create relatively ideal detection conditions to reliably identify the customized target. When the road traffic volume is large, the complexity of the sensing scene increases, and it will be difficult to reliably identify the customized target. When the road side sensor only outputs target-level vehicle sensing data, it will be particularly difficult to identify the vehicle trajectory carrying the positioning device in the multi-target trajectory data output by the road side sensing device. In order to reliably realize parameter calibration, it is often necessary to complete the calibration work under the condition of road closure or when the traffic volume is small.

[0005] However, in the context of a large number of road side sensing devices, the sensing devices need to be recalibrated in time and space parameters in the case of first installation, damage and replacement, vibration and insecure installation, etc. It is time-consuming and laborious to carry out calibration work using the traditional method. At the same time, in the application of vehicle-road cooperation, the real-time and accuracy requirements of road side sensing data are continuously improving, and the sensing failure caused by the untimely update of the space-time parameters of the road side sensor will further cause decision-making failure and safety hazards.

[0006] Therefore, it is urgent to study a roadside multi-sensor space-time synchronization method to realize online automatic calibration of space-time parameters of roadside sensors. SUMMARY

[0007] The present application aims to overcome the defects of the prior art and provides a multi-sensor space-time calibration method for a vehicle-road cooperative system based on graph matching.

[0008] The object of the present application can be achieved by the following technical solutions:

[0009] A multi-sensor space-time calibration method for a vehicle-road cooperative system based on graph matching, the method comprising the following steps:

[0010] S1, the networked vehicle passes through the roadside multi-sensor sensing range, collects the self-positioning data of the networked vehicle and its perception data of the surrounding traffic flow, collects the perception data of the roadside multi-sensor for the traffic flow, and processes all the data into multi-target trajectory data, including the vehicle position, vehicle speed and vehicle ID of the vehicle target in the respective coordinate system;

[0011] S2, using a graph matching method to match the vehicle ID collected by the roadside multi-sensor, identifying the networked vehicle trajectory in the multi-target trajectory data collected by the roadside multi-sensor;

[0012] S3, using a Gauss-Newton method to solve the space-time parameters from the networked vehicle positioning data coordinate system to the roadside multi-sensor data coordinate system, so that the networked vehicle trajectory collected by the roadside multi-sensor is aligned with the networked vehicle self-positioning trajectory in a unified space-time coordinate system;

[0013] S4, according to the space-time parameters of each roadside multi-sensor obtained in S3, transforming the roadside multi-sensor collected data to realize the space-time synchronization of the roadside multi-sensor.

[0014] Further, S2 specifically comprises the following steps:

[0015] S201, for each segment of trajectory data, a continuous time state estimation model is established using a Gaussian process to query the position and speed of the target at any time point between the appearance and disappearance of the vehicle target;

[0016] S202, according to the vehicle-road data system time delay and the continuous time state estimation model, the multi-target trajectory data collected by the roadside multi-sensor is interpolated at the time stamp of the vehicle end perception data after adding the vehicle-road data system time delay compensation;

[0017] S203, defining a weighted graph sequence to describe the detection data of a single sensor in a period of time, storing the features of a single connected vehicle with different vehicle targets as nodes, and storing the features between two connected vehicles on the connection between nodes;

[0018] S204, establishing a vehicle-side perception target feature map with the self-positioning data of the connected vehicle and the perception data of the surrounding traffic flow in a period of time, and establishing a roadside perception target feature map according to the perception data of the traffic flow of the roadside multi-sensor at the same time;

[0019] S205, calculating the similarity matrix between the vehicle-side perception target feature map and the roadside perception target feature map, calculating the confidence degree of the node matching relationship according to the node matching relationship, and searching for the node matching relationship between the vehicle-side perception target feature map and the roadside perception target feature map and the vehicle-road data system delay;

[0020] S206, repeating S202-S205 until the confidence degree of the node matching relationship converges to a maximum value;

[0021] S207, according to the obtained node matching relationship, screening out the connected vehicle trajectory from the multi-target trajectory data collected by the roadside multi-sensor, and obtaining a rough estimation of the vehicle-road data system delay according to the search result of the vehicle-road data system delay.

[0022] Further, the features of the single connected vehicle include vehicle position and vehicle speed.

[0023] Further, the features between two connected vehicles include vehicle relative position and vehicle relative speed.

[0024] Further, S201 specifically includes:

[0025] S2011, defining a continuous-time Gaussian process by a time-varying differential equation;

[0026] S2012, calculating the target prior state mean and covariance at the interpolation time and the observation time according to the defined Gaussian process from the initial state and the covariance;

[0027] S2013, calculating the target posterior state estimate at the observation time from the observation value, the observation noise, the target prior state mean and covariance at the observation time, and the maximum likelihood estimation;

[0028] S2014, calculating the target posterior state estimate at the interpolation time from the target prior state and target posterior state estimate values at adjacent observation times and the target prior state estimate at the interpolation time.

[0029] Further, the target prior state mean and covariance are represented as:

[0030]

[0031] wherein, denotes the target prior state mean, denotes the covariance, , is a known initial state vector and covariance matrix, is a system matrix, is a process noise power matrix, is a state transition matrix from time to time is a transpose matrix of the state transition matrix from time to time is a transpose matrix of the state transition matrix from time to time is a state transition matrix from time to time is a state transition matrix from time to time is a state transition matrix from time to time is calculated under the uniform velocity model as shown below:

[0032] .

[0033] Further, the searching of the node matching relationship between the vehicle-end perception target feature map and the roadside perception target feature map and the time delay of the vehicle-road data system in S205 is specifically:

[0034] S2051, a gradually non-convex and concave process is used to search the node matching relationship between the vehicle-end perception target feature map and the roadside perception target feature map under a fixed time delay, so that the confidence degree of the matching relationship is maximum.

[0035] S2052, a parabolic interpolation and a golden section method are used to update the search parameters of the time delay, so that the confidence degree of the output matching relationship under the time delay is maximum under different time delays.

[0036] Further, S2052 is specifically: under a fixed time delay , the search parameters of the time delay are initialized, wherein the upper and lower bounds of the search interval are , and the historical suboptimal value is ; if satisfies the parabolic interpolation condition, the parabolic interpolation is used to update the time delay ; otherwise, the golden section method is used to update according to the upper and lower bounds of the search interval , and finally the search parameters are updated.

[0037] Further, the S3 is specifically:

[0038] S301, defining a networking vehicle positioning data coordinate system as a source coordinate system, a road side multi-sensor data coordinate system as a target coordinate system, and defining a transformation parameter from the source coordinate system to the target coordinate system;

[0039] S302, taking the sampling time of the road side sensor as a reference, and interpolating the vehicle end positioning trajectory data after joining the vehicle-road data system delay compensation;

[0040] S303, calculating the projection of the vehicle end positioning trajectory data in the road side multi-sensor data coordinate system;

[0041] S304, calculating the deviation of the projection of the vehicle end positioning trajectory data in the road side multi-sensor data coordinate system and the road side multi-sensor collected networking vehicle trajectory data;

[0042] S305, updating the space-time parameters of the road side sensor by using the Gauss-Newton method;

[0043] S306, repeating S302-S305 until the deviation of the projection of the vehicle end positioning trajectory data in the road side multi-sensor data coordinate system and the road side multi-sensor collected networking vehicle trajectory data converges to a minimum value;

[0044] S307, outputting the space-time parameters of the road side multi-sensor.

[0045] Further, the transformation parameter in S301 includes a time parameter and a space parameter.

[0046] Compared with the prior art, the present application has the following beneficial effects:

[0047] 1. The present application is based on target level vehicle-road perception data, and the information of the networking vehicle sensing the surrounding traffic flow is used as the basis for vehicle-road target matching, and the high-precision positioning data of the networking vehicle is used as the reference for the calibration of the road side sensor. The space-time parameters of the road side sensor are automatically calibrated, the parameter calibration of the road side multi-sensor is completed quickly and automatically, and the space-time synchronization of the road side multi-sensor is efficiently realized.

[0048] 2. The present application can calibrate the parameters without affecting the traffic, saving time and effort, and having high real-time performance and accuracy. It can effectively prevent the perception failure caused by the untimely update of the space-time parameters of the road side multi-sensor, and effectively reduce the safety hazards and decision-making errors. BRIEF DESCRIPTION OF DRAWINGS

[0049] Figure 1 The vehicle ID matching flowchart of the road side multi-sensor and the networking vehicle of the present application;

[0050] Figure 2A flow chart of a roadside multi-sensor parameter calibration of the present application;

[0051] Figure 3 A correspondence diagram of a networked vehicle positioning data coordinate system and a roadside multi-sensor data coordinate system of the present application;

[0052] Figure 4 A second-order similarity matrix diagram of a vehicle-end perception target feature map and a roadside perception target feature map of the present application;

[0053] Figure 5 A roadside millimeter wave radar calibration result diagram of the present application. DETAILED DESCRIPTION

[0054] The present application will be described in detail below with reference to the accompanying drawings and specific embodiments. The embodiments are implemented on the premise of the technical solutions of the present application, and detailed implementation modes and specific operation processes are given, but the protection scope of the present application is not limited to the following embodiments.

[0055] EMBODIMENT

[0056] The present embodiment is based on millimeter wave radar data of a Donghai Bridge traffic monitoring system and perception data of an intelligent truck. In the present embodiment, the millimeter wave radar data includes the position and speed of a vehicle target in a radar coordinate system, and the sampling frequency is 10 Hz; the intelligent truck positioning data and perception data have a frequency of 25 Hz, and are asynchronous with the roadside millimeter wave radar in sampling time.

[0057] A multi-sensor space-time calibration method for a vehicle-road cooperative system based on graph matching, the method comprising the following steps:

[0058] S1, a networked vehicle passes through a roadside multi-sensor perception range, collects self-positioning data of the networked vehicle and its perception data of surrounding traffic flow, collects perception data of the traffic flow by the roadside multi-sensor, and processes all the data into multi-target trajectory data, including vehicle position, vehicle speed and vehicle ID of the vehicle target in respective coordinate systems.

[0059] S2, as shown in the figure, a graph matching method is used to match vehicle IDs collected by the roadside multi-sensor, and the trajectory of the networked vehicle is identified in the multi-target trajectory data collected by the roadside multi-sensor. Figure 1

[0060] S201, for each piece of trajectory data, a continuous time state estimation model is established by using a Gaussian process to query the position and speed of the target at any time point between the appearance and disappearance of the vehicle target; for each piece of trajectory data, a continuous time Gaussian process is established with a discrete time observation model , wherein, ​respectively the target state prior mean and covariance function, is the target state at the th sampling time, is the target state at the th sampling time, is the measurement at the th sampling time, is the observation matrix, is the measurement noise satisfying where

[0061] is the covariance matrix of the measurement noise; is the system matrix, is the input signal, is a Gaussian process with zero mean and covariance function defined by the process noise power matrix and the Dirac function ;

[0062] S2012, according to the defined Gaussian process, the target prior state mean and covariance at the interpolation time and observation time are calculated from the initial state and covariance, as shown in the following formula:

[0063]

[0064] where, , is the known initial state vector and covariance matrix, is the state transition matrix from the th time to the th time, is the transpose matrix of the state transition matrix from the th time to the th time, is the transpose matrix of the state transition matrix from the th time to the th time, is the state transition matrix from the th time to the th time, under the uniform speed model, the following is calculated;

[0065]

[0066] the target prior state mean and covariance at the observation time are calculated as shown below:

[0067] ​​

[0068] wherein,

[0069]

[0070]

[0071] S2013, calculating the target posterior state estimation at the observation time by maximum likelihood estimation from the observation value, observation noise, target prior state mean value and covariance at the observation time;

[0072] S2014, calculating the target posterior state estimation at the interpolation time from the target prior state and target posterior state estimation values at adjacent observation times and the target prior state estimation at the interpolation time as follows;

[0073]

[0074] wherein, only adjacent observation times of the target prior state , , and the target posterior state , are simple combinations;

[0075] S202, according to the vehicle-road data system delay and the continuous time state estimation model, adding the vehicle-road data system delay compensation to the multi-target trajectory data collected by the roadside multi-sensor at the timestamp of the vehicle-end perception data, and then interpolating;

[0076] S203, defining a weighted graph sequence , describing the detection data of a single sensor in a period of time, storing the features of a single connected vehicle with different vehicle targets as nodes, and storing the features between two connected vehicles on the connection between nodes; the features of a single connected vehicle include vehicle position and vehicle speed; the features between two connected vehicles include vehicle relative position and vehicle relative speed; wherein is the set of all target numbers observed by the sensor in this period, and the total number of observed targets is , is the set of observation times, and the total number of frames is , the frame corresponds to the observation time :

[0077]

[0078] The features of a single connected vehicle, i.e. first-order features, Store the features between each pair of connected vehicles, i.e. second-order features, calculate first-order features With second-order features As follows:

[0079]

[0080] Wherein, Indicates the target Speed vector at Time, Indicates the target Position vector at Time, Indicates the second-order norm of the vector, Speed normalization coefficient, Distance normalization coefficient; remove part of the edges in the graph sequence, reduce the density of node connection as follows:

[0081]

[0082] Wherein, Indicates the target Time interval of appearance, Indicates the target Length of appearance time, Time overlap threshold, Distance threshold, The graph in the Frame of the graph sequence with the target And the target As the vertex, the effectiveness is judged according to the time overlap, The effectiveness is judged according to the spatial distance, The second-order feature matrix after filtering out invalid edges;

[0083] S204, establish the vehicle-end perception target feature graph with the self-positioning data of the connected vehicle in a period of time and its perception data of the surrounding traffic flow, and establish the road-side perception target feature graph according to the perception data of the traffic flow of the road-side multi-sensor interpolated at the same time;

[0084] S205, calculate the similarity matrix between the vehicle-end perception target feature graph and the road-side perception target feature graph, calculate the confidence degree of the matching relationship according to the node matching relationship, and search the node matching relationship of the vehicle-end perception target feature graph and the road-side perception target feature graph and the time delay of the vehicle-road data system, specifically:

[0085] The data of the vehicle-end sensor In a period of time is constructed into a graph sequence , and Is used as the criterion to join the time offset interpolate the data of the roadside sensors , construct the sequence of maps , the vehicle-end sensors , the roadside sensors respectively detect , targets; define represent second-order matching relationships between , the targets detected by the vehicle-end sensors match the targets observed by the roadside sensors respectively, all possible second-order matching relationships constitute a set , on this basis, add a set of pseudo-targets to the detected targets of the roadside sensors as a substitute matching of the missed targets when the roadside sensors miss detection, which can be regarded as a case of matching loss; all possible second-order substitute matching relationships constitute a set , including three cases: the starting node mismatch , the ending node mismatch and both endpoints mismatch ; the similarity of first-order and second-order features between is measured by using a Gaussian kernel function, and for the feature similarity in the case of matching loss, a fixed value is used to define the maximum allowed matching error. The similarity matrix of the vehicle-road target feature map is calculated as follows: wherein is the maximum allowed deviation of the first-order feature, is the maximum allowed deviation of the second-order feature, and the second-order similarity matrix of the vehicle-end perception target feature map and the roadside perception target feature map is shown in .

[0086]

[0087] wherein is the maximum allowed deviation of the first-order feature, is the maximum allowed deviation of the second-order feature, and the second-order similarity matrix of the vehicle-end perception target feature map and the roadside perception target feature map is shown in Figure 4 .

[0088] Normalize the similarity matrix, first average the similarity matrix at each time in the observation period in time:

[0089]

[0090] Then, the time-averaged similarity matrix is processed by the weight between the first-order and second-order feature similarities:

[0091] ​​​​

[0092] obtain a similarity matrix between the final vehicle-end perception target feature map and the roadside perception target feature map ;

[0093] S2051, establish a vehicle-road target matching relationship optimization model:

[0094]

[0095]

[0096]

[0097] wherein, the optimization variable is a 01 vector representing the matching relationship, is the number of targets detected by the vehicle-end sensor , the sum of the number of targets detected by the vehicle-end sensor , the roadside sensor (contains artificially added pseudo targets); in the objective function is the vehicle-road target matching similarity matrix under the time delay ; under the fixed time delay , the node matching relationship between the vehicle-end perception target feature map and the roadside perception target feature map under the fixed time delay is searched by using a gradually non-convex and concave process, so that the confidence degree of the matching relationship is maximum, and specifically: Set the gradual coefficient

[0098] and the initial value of the relaxed matching relationship , such as:

[0099] ,

[0100] According to the Hungarian algorithm, the optimization direction is solved, such as:

[0101]

[0102] According to the line search, the iteration rate is solved, such as:

[0103]

[0104] Update the relaxed matching relationship, such as:

[0105]

[0106] When is not about ​Converge, continue to solve the optimization direction and the iteration rate ; update the progressive coefficient, such as: ; when , continue to solve the optimization direction and the iteration rate ; output and

[0107] wherein,

[0108]

[0109]

[0110] ;

[0111] S2052, the search parameters of the time delay are updated by using parabolic interpolation and the golden section method, so that the confidence degree of the matching relationship output under the time delay is maximum under different time delays; under the fixed time delay , the search parameters of the time delay are initialized, wherein the upper and lower bounds of the search interval are , and the historical suboptimal value is ; if the parabolic interpolation condition is met, the time delay is updated by using parabolic interpolation; otherwise, the golden section method is used to update according to the upper and lower bounds of the search interval , and finally the search parameters are updated;

[0112] S206, S202-S205 are repeated until the confidence degree of the node matching relationship converges to a maximum value;

[0113] S207, according to the obtained node matching relationship, the networking vehicle trajectory is screened out from the multi-target trajectory data collected by the roadside multi-sensor, and according to the time delay search result of the vehicle-road data system, a rough estimation of the time delay of the vehicle-road data system is obtained.

[0114] S3, as shown in Figure 2 , the Gaussian-Newton method is used to solve the space-time parameters from the networking vehicle positioning data coordinate system to the roadside multi-sensor data coordinate system, so that the networking vehicle trajectory collected by the roadside multi-sensor is aligned with the networking vehicle self-positioning trajectory in a unified space-time coordinate system.

[0115] S301, as shown in Figure 3 , the networking vehicle positioning data coordinate system is defined as the source coordinate system and is marked as , and the roadside multi-sensor data coordinate system is defined as the target coordinate system and is marked as , define the transformation parameters from the source coordinate system to the target coordinate system; the transformation parameters include time parameters and space parameters; for the same target motion trajectory, the coordinates in the source coordinate system , the coordinates in the target coordinate system , define the to-be-calibrated parameters , wherein is a rotation angle, , is a translation amount, is a time delay;

[0116] S302, taking the sampling time of the roadside sensor as a reference, interpolating the vehicle-end positioning trajectory data after adding the delay compensation of the vehicle-road data system to obtain :

[0117]

[0118] S303, calculating the projection of the vehicle-end positioning trajectory data in the roadside multi-sensor data coordinate system , the calculation formula is as follows:

[0119]

[0120] , wherein is a rotation matrix, is a translation vector:

[0121]

[0122] S304, calculating the deviation between the projection of the vehicle-end positioning trajectory data in the roadside multi-sensor data coordinate system and the trajectory data of the connected vehicle collected by the roadside multi-sensor;

[0123] S305, updating the space-time parameters of the roadside sensor by using the Gauss-Newton method;

[0124] S306, repeating S302-S305 until the deviation between the projection of the vehicle-end positioning trajectory data in the roadside multi-sensor data coordinate system and the trajectory data of the connected vehicle collected by the roadside multi-sensor converges to a minimum value;

[0125] S307, outputting the space-time parameters of the roadside sensor.

[0126] establish a nonlinear least squares model:

[0127]

[0128]

[0129] ​

[0130] wherein is the parameter to be calibrated, is the projection deviation of each measurement point; the time delay estimation output by the matching algorithm is used as the initial value of the time offset If the above value is not available, the time delay initial value can be set by the time center alignment method as follows:

[0131]

[0132] In the case where the time correspondence is determined by the time delay, the singular value decomposition method is used to solve the initial value of the rotation matrix :

[0133]

[0134]

[0135]

[0136] Under the initial value of the rotation matrix , the mean of the translation amount of each point is used as the translation initial value :

[0137]

[0138] The Jacobian matrix is calculated:

[0139]

[0140] According to iterative calculation until convergence to obtain the space-time parameters to be calibrated, wherein:

[0141]

[0142]

[0143] S4, according to the space-time parameters of each roadside multi-sensor obtained in S3, the data collected by the roadside multi-sensor is transformed to realize the space-time synchronization of the roadside multi-sensor. According to the calibrated space-time parameters, the data of multiple roadside sensors can be projected into a unified plane coordinate system to realize the space-time synchronization of multiple sensors. In the embodiment, the calibration result of a roadside millimeter wave radar is shown in Figure 5 The projection of the vehicle positioning data and the millimeter wave radar trajectory data fits well, and the root mean square error is 0.09m.

[0144] The preferred embodiments of the present application have been described above in detail. It should be understood that modifications and variations to the present application can be affected by those skilled in the art without departing from the scope of the application. Accordingly, it is intended that all possible modifications and alterations be included within the scope of the present application as defined by the following claims.

Claims

1. A multi-sensor space-time calibration method based on graph matching for a vehicle infrastructure integration system, characterized in that, The method comprises the following steps: S1, the networking vehicle passes through the roadside multi-sensor sensing range, collects the self-positioning data of the networking vehicle and its sensing data of the surrounding traffic flow, collects the sensing data of the roadside multi-sensor for the traffic flow, and processes all the data into multi-target trajectory data, including the vehicle position, vehicle speed and vehicle ID of the vehicle target in the respective coordinate system; S2, the vehicle ID collected by the roadside multi-sensor is matched by using a graph matching method, and the networking vehicle trajectory is identified in the multi-target trajectory data collected by the roadside multi-sensor; S3, the spatio-temporal parameters from the networking vehicle positioning data coordinate system to the roadside multi-sensor data coordinate system are solved by using a Gauss-Newton method, so that the networking vehicle trajectory collected by the roadside multi-sensor is aligned with the self-positioning trajectory of the networking vehicle in a unified spatio-temporal coordinate system; S4, according to the spatio-temporal parameters of each roadside multi-sensor obtained in S3, the roadside multi-sensor collected data is transformed to realize the spatio-temporal synchronization of the roadside multi-sensor. 2.The multi-sensor space-time calibration method based on graph matching for vehicle infrastructure integration system according to claim 1, wherein, The S2 comprises the following steps: S201, for each segment of trajectory data, a continuous time state estimation model is established by using a Gaussian process to query the position and speed of the target at any time point between the appearance and disappearance of the vehicle target; S202, according to the vehicle-road data system time delay and the continuous time state estimation model, the multi-target trajectory data collected by the roadside multi-sensor is interpolated at the time stamp of the vehicle end sensing data after adding the vehicle-road data system time delay compensation; S203, a weighted graph sequence is defined to describe the detection data of a single sensor within a period of time, different vehicle targets are used as nodes to store the characteristics of a single networking vehicle, and the characteristics between two networking vehicles are stored on the connection between the nodes; S204, a vehicle end sensing target feature graph is established according to the self-positioning data of the networking vehicle and its sensing data of the surrounding traffic flow within a period of time, and a roadside sensing target feature graph is established according to the interpolated traffic flow sensing data of the roadside multi-sensor at the synchronization time; S205, the similarity matrix between the vehicle end sensing target feature graph and the roadside sensing target feature graph is calculated, the confidence degree of the matching relationship is calculated according to the node matching relationship, and the node matching relationship of the vehicle end sensing target feature graph and the roadside sensing target feature graph and the vehicle-road data system time delay are searched; S206, S202-S205 are repeated until the confidence degree of the node matching relationship converges to a maximum value; S207, according to the obtained node matching relationship, the networking vehicle trajectory is screened out in the multi-target trajectory data collected by the roadside multi-sensor, and a rough estimation of the vehicle-road data system time delay is obtained according to the search result of the vehicle-road data system time delay.

3. The multi-sensor space-time calibration method based on graph matching for vehicle infrastructure cooperation system according to claim 2, characterized in that, The characteristics of the single networking vehicle include vehicle position and vehicle speed.

4. The multi-sensor space-time calibration method based on graph matching for vehicle infrastructure cooperation system according to claim 2, characterized in that, The characteristics between two networking vehicles include vehicle relative position and vehicle relative speed.

5. The multi-sensor space-time calibration method based on graph matching for vehicle infrastructure integration system according to claim 2, wherein, The S201 is specifically: S2011, a continuous time Gaussian process is defined by a time-varying differential equation; S2012, according to the defined Gaussian process, the target prior state mean and covariance at the interpolation time and observation time are calculated from the initial state and covariance. S2013, calculating the target posterior state estimation at the observation time by maximum likelihood estimation from the observation value, observation noise, target prior state mean value and covariance at the observation time; S2014, calculating the target posterior state estimation at the interpolation time from the target prior state estimation and target posterior state estimation at the adjacent observation time and target prior state estimation at the interpolation time.

6. The multi-sensor space-time calibration method based on graph matching for vehicle infrastructure cooperative systems according to claim 5, wherein, The target prior state mean value and covariance are expressed as: wherein, represents a target prior state mean, represents a covariance, is a known initial state vector and covariance matrix, L is a system matrix, Q c is a process noise power matrix, Φ(t, t1) is a state transition matrix from t time to t1 time, Φ(t ′ ,t1) T is a transpose matrix of the state transition matrix from t ′ time to t1 time, Φ(t ′ ,s) T is a transpose matrix of the state transition matrix from t ′ time to s time, Φ(t, s) is a state transition matrix from s time to t time, Φ(t, s) is calculated as shown below under a uniform speed model:

7. The multi-sensor space-time calibration method based on graph matching for vehicle infrastructure cooperative systems according to claim 2, wherein, The node matching relationship between the vehicle-end perception target feature map and the roadside perception target feature map and the vehicle-road data system time delay in S205 are searched as follows: S2051, searching the node matching relationship between the vehicle-end perception target feature map and the roadside perception target feature map under a fixed time delay by using a gradually non-convex and concave process, so that the confidence degree of the matching relationship is maximum; S2052, updating the search parameters of the time delay by using parabolic interpolation and golden section method, so that the confidence degree of the output matching relationship under the time delay is maximum under different time delays.

8. The multi-sensor space-time calibration method based on graph matching for a vehicle infrastructure cooperative system according to claim 7, wherein, The S2052 is specifically: in fixed time delay t d Next, the search parameters of the initialization time delay are updated, wherein the upper and lower bounds of the search interval are a, b, the historical suboptimal values are v, w, and t d If the parabolic interpolation condition is met, the parabolic interpolation is used to update the time delay t d ; otherwise, the golden section method is used to update t d according to the upper and lower bounds a, b of the search interval, and finally the search parameters a, b, v, and w are updated.

9. The multi-sensor space-time calibration method based on graph matching for vehicle infrastructure cooperative systems according to claim 1, wherein, S3 is specifically as follows: S301, defining the networking vehicle positioning data coordinate system as a source coordinate system and the roadside multi-sensor data coordinate system as a target coordinate system, and defining the transformation parameters from the source coordinate system to the target coordinate system; S302, taking the sampling time of the roadside sensor as a reference, and interpolating the vehicle-end positioning trajectory data after adding the vehicle-road data system time delay compensation; S303, calculating the projection of the vehicle-end positioning trajectory data in the roadside multi-sensor data coordinate system; S304, calculating the deviation between the projection of the vehicle-end positioning trajectory data in the roadside multi-sensor data coordinate system and the networking vehicle trajectory data collected by the roadside multi-sensor; S305, updating the space-time parameters of the roadside sensor by using the Gauss-Newton method; S306, repeating S302-S305 until the deviation between the projection of the vehicle-end positioning trajectory data in the roadside multi-sensor data coordinate system and the networking vehicle trajectory data collected by the roadside multi-sensor converges to a minimum value; S307, outputting the space-time parameters of the roadside sensor.

10. The multi-sensor space-time calibration method based on graph matching for a vehicle infrastructure cooperative system according to claim 9, wherein, The transformation parameters in S301 include time parameters and space parameters.

Images