Map data system fusing two or more versions based on space Kalman filtering

Abstract

A system for fusing two or more versions of map data together includes one or more central computers that receive road network data representing a road network of a predefined geo-fenced area. A central computer receives road network data including discrete random curves representing lane markings. The discrete random curve includes a plurality of state vectors, each state vector defined by a respective position and tangent angle. The central computer estimates the position of a state vector of the discrete random curve based on the signed distance and the tangent angle by minimizing a space Kalman filtering cost function, and executes a Kalman smoothing function to estimate the position and the tangent angle of the state vector, the position and the tangent angle of the state vector being part of the discrete random curve, the state vectors each represent a map point of the fused map data.

Description

Technical Field

[0001] The present disclosure relates to a system for fusing two or more versions of map data together to create fused map data based on spatial Kalman filtering. Background Art

[0002] Autonomous vehicles perform various tasks, such as, but not limited to, perception, localization, mapping, path planning, decision making, and motion control. Autonomous vehicles rely on map data to perform many tasks, such as localization, mapping, and path planning. It should be understood that different versions of map data representing the same geographic area may be generated, where each version of the map data is generated from a different data source.

[0003] An example of a version of map data is crowdsourced map data. Crowdsourced map data is generated based on Global Positioning System (GPS) data and data collected by vehicles on the road. Some examples of data collected by vehicles include image data collected by onboard cameras, vehicle speed, and yaw sensor data. While crowdsourced maps provide real-time information for the drawing of lane markings and clear lane line attributes, crowdsourced maps can face challenges due to calibration issues of onboard cameras. In addition, GPS data may introduce bias and random noise to crowdsourced maps.

[0004] Map data based on telemetry data, such as high-speed vehicle telemetry (HSVT) sources, can provide high confidence and are generally relatively easy to obtain. However, telemetry-based map data does not provide lane attributes, such as color and type. In addition, telemetry-based map data is determined based on the behavior of individual drivers operating the vehicle. Aerial map data based on satellite imagery data can provide high accuracy, however, aerial map data may not consistently provide lane attributes. In addition, aerial maps are time-consuming to create, may be out of date (depending on when the images were collected), and may contain areas that are obscured due to issues such as tree cover. Finally, high-definition map data is generated based on data collected from survey vehicles, which can also provide high accuracy as well as lane attributes. However, updating high-definition map data is time-consuming.

[0005] In one approach to mitigate the above challenges, map data can be combined based on a cascaded hierarchical architecture. In this approach, if a particular version of map data does not provide the level of accuracy required by a particular application, the cascaded hierarchical architecture will provide another version of map data instead. However, the cascaded hierarchical architecture only uses each version of map data as an alternative option and does not integrate different versions of map data together. Therefore, although maps for autonomous vehicles achieve their intended purpose, there is still a need in the art for an improved method for obtaining map data to mitigate the above challenges. Summary of the invention

[0006] According to several aspects, a system for fusing two or more versions of map data together to create fused map data is disclosed. The system includes one or more central computers storing two or more versions of map data and ground truth map data. Each version of the map data represents a predefined geo-fenced area. The one or more central computers execute instructions to receive road network data representing a road network of the predefined geo-fenced area, a first set of map data points, and a second set of map data points, wherein the road network data includes a discrete random curve representing a lane marking, wherein the discrete random curve includes a plurality of state vectors, each state vector being defined by a respective position and a tangent angle. The central computer determines a signed distance, wherein the signed distance includes a first perpendicular distance measured from a corresponding first map data point to the discrete random curve and a second perpendicular distance measured from a corresponding second map data point to the discrete random curve. The one or more central computers estimate the position of the state vector of the discrete random curve based on the signed distance and the tangent angle by minimizing a spatial Kalman filter cost function. The central computer executes a Kalman smoothing function to estimate a position and a tangent angle of a state vector that is a portion of a discrete random curve, wherein the state vectors each represent a map point of the fused map data.

[0007] In another aspect, the discrete random curve is one of: a polygonal curve and a Markov random curve.

[0008] In yet another aspect, one or more central computers execute instructions to detect a data point representing a new state vector introduced into the discrete random curve, and in response to detecting the data point, estimate an updated state vector, wherein the updated state vector indicates the position of the new data point and the corresponding value of the tangent angle.

[0009] In one aspect, the updated state vector is represented as:

[0010]

[0011] in, represents the updated state vector, k represents the total number of state vectors, represents the predicted state, d k represents a matrix indicating the horizontal distance and the vertical distance of the first map data point and the horizontal distance and the vertical distance of the second map data point, μ v represents the deviation of the measurement noise, represents the covariance of the measurement noise, represents the covariance matrix.

[0012] On the other hand, the spatial Kalman filter cost function is expressed as:

[0013]

[0014] in, Represents the updated state vector s k The estimated position of k represents the total number of state vectors, d i represents the signed distance between the first map data point and the second map data point, μ v represents the deviation of the measurement noise, q u represents the variance of additive white Gaussian noise, represents the covariance of the measurement noise, φ i Represents the tangent angle.

[0015] In yet another aspect, one or more central computers execute instructions to receive ground truth map data, execute one or more machine learning algorithms to determine a first deviation and a first variance for a first set of map data points and a second deviation and a second variance for a second set of map data points, and execute an iterative algorithm to remove outliers from the first map data points, the second map data points, and the ground truth map data.

[0016] In one aspect, the covariance of the measurement noise is determined based on ground truth map data.

[0017] In another aspect, the iterative algorithm is a Randomized Sample Consensus (RANSAC) voting algorithm.

[0018] In yet another aspect, the deviation of the measurement noise is determined based on a first deviation of the first set of map data points and a second deviation of the second set of map data points.

[0019] In one aspect, the Kalman smoothing function is expressed as:

[0020]

[0021] in, Represents the updated state vector s k The estimated position of N represents the total number of state vectors plus one, d i represents the signed distance between the first map data point and the second map data point, μ v represents the deviation of the measurement noise, q u represents the variance of additive white Gaussian noise, represents the covariance of the measurement noise, φ i Represents the tangent angle.

[0022] In another aspect, the two or more versions of map data are each based on one or more of: global positioning system (GPS) data, perception data, high speed vehicle telemetry (HSVT) sources, satellite imagery data, and data collected by a survey vehicle.

[0023] In yet another aspect, a method for fusing two or more versions of map data is disclosed. The method includes receiving, by one or more central computers, road network data representing a road network of a predefined geofenced area, a first set of map data points, and a second set of map data points, wherein the road network data includes a discrete random curve representing a lane marking, wherein the discrete random curve includes a plurality of state vectors, each state vector being defined by a respective position and a tangent angle. The method includes determining, by the one or more central computers, a signed distance, wherein the signed distance includes a first perpendicular distance measured from a corresponding first map data point to the discrete random curve and a second perpendicular distance measured from a corresponding second map data point to the discrete random curve. The method includes estimating, by the one or more central computers, a position of a state vector of the discrete random curve based on the signed distance and the tangent angle by minimizing a spatial Kalman filter cost function. The method also includes executing a Kalman smoothing function to estimate a position and a tangent angle of a state vector, the position and the tangent angle of the state vector being a portion of the discrete random curve, wherein the state vectors each represent a map point of the fused map data.

[0024] On the other hand, the method also includes detecting, by one or more central computers, a data point representing a new state vector introduced into the discrete random curve, and estimating an updated state vector in response to detecting the data point, wherein the updated state vector indicates the position of the new data point and the corresponding value of the tangent angle.

[0025] In one aspect, a system for fusing two or more versions of map data together to create fused map data is disclosed. The system includes one or more central computers storing two or more versions of map data and ground truth map data, wherein each version of the map data represents a predefined geo-fenced area. The one or more central computers execute instructions to receive road network data representing a road network of the predefined geo-fenced area, a first set of map data points, and a second set of map data points, wherein the road network data includes a discrete random curve representing a lane marking, wherein the discrete random curve includes a plurality of state vectors, each state vector being defined by a respective position and a tangent angle. The central computer determines a signed distance, wherein the signed distance includes a first perpendicular distance measured from a corresponding first map data point to the discrete random curve and a second perpendicular distance measured from a corresponding second map data point to the discrete random curve. The one or more central computers estimate a position of a state vector of the discrete random curve based on the signed distance and the tangent angle by minimizing a spatial Kalman filter cost function, wherein the spatial Kalman filter cost function is expressed as:

[0026]

[0027] in, represents the updated state vector s k The estimated position of k represents the total number of state vectors, d i represents the signed distance between the first map data point and the second map data point, μ v represents the deviation of the measurement noise, q u represents the variance of additive white Gaussian noise, represents the covariance of the measurement noise, φ i One or more central computers execute a Kalman smoothing function to estimate a position and a tangent angle of a state vector, the position and the tangent angle of the state vector being a portion of a discrete random curve, wherein the state vectors each represent a map point of the fused map data.

[0028] In another aspect, one or more central computers execute instructions to receive ground truth map data, execute one or more machine learning algorithms to determine a first deviation and a first variance for a first set of map data points and a second deviation and a second variance for a second set of map data points, and execute an iterative algorithm to remove outliers from the ground truth map data.

[0029] In yet another aspect, the covariance of the measurement noise is determined based on ground truth map data.

[0030] In one aspect, the iterative algorithm is a Randomized Sample Consensus (RANSAC) voting algorithm.

[0031] In another aspect, one or more central computers execute instructions to detect a data point representing a new state vector introduced into the discrete random curve, and in response to detecting the data point, estimate an updated state vector, wherein the updated state vector indicates the position of the new data point and the corresponding value of the tangent angle.

[0032] In yet another aspect, the updated state vector is represented as:

[0033]

[0034] in, represents the updated state vector, represents the predicted state, d k represents a matrix indicating the lateral distance and the longitudinal distance of the first map data point and the lateral distance and the longitudinal distance of the second map data point, represents the covariance matrix.

[0035] In one aspect, the Kalman smoothing function is expressed as:

[0036]

[0037] in, Represents the updated state vector s k The estimated position of , N represents the total number of state vectors plus one.

[0038] Further areas of applicability will become apparent from the description provided herein.It should be understood that the description and specific examples are intended for purposes of illustration only and are not intended to limit the scope of the present disclosure. BRIEF DESCRIPTION OF THE DRAWINGS

[0039] The drawings described herein are for illustration purposes only and are not intended to limit the scope of the present disclosure in any way.

[0040] Figure 1 is a schematic diagram of a disclosed system for fusing two or more versions of map data together according to an exemplary embodiment, wherein the system includes one or more central computers that receive map data from different sources over a network;

[0041] Figure 2 is a diagram showing an exemplary embodiment of the present invention. Figure 1 a block diagram of the software architecture of the central computer or computers shown;

[0042] Figure 3 is an exemplary illustration of a portion of a road network including discrete random curves representing lane markings according to an exemplary embodiment;

[0043] Figure 4 According to an exemplary embodiment, Figure 3 A portion of the discrete random curve shown; and

[0044] Figure 5 A process flow diagram illustrating a method for providing a method for providing a Figure 1 The system shown fuses two or more versions of map data together. DETAILED DESCRIPTION

[0045] The following description is merely exemplary in nature and is not intended to limit the present disclosure, application, or uses.

[0046] Reference Figure 1, an exemplary system 10 for fusing two or more versions of map data together is shown. The system 10 includes one or more central computers 20 located in a back-end office 22, wherein the one or more central computers 20 wirelessly communicate with one or more communication networks 24. The one or more central computers 20 obtain two or more versions of map data and ground-truth map data through the one or more communication networks 24. In an embodiment, the one or more central computers 20 store the two or more versions of map data and ground-truth map data in a local memory. Each version of the map data represents the same predefined geo-fenced area, wherein the predefined geo-fenced area represents a real-world geographic area defined by a virtual perimeter.

[0047] It should be understood that each version of the map data is based on a unique data source, such as global positioning system (GPS) data, perception data, high-speed vehicle telemetry (HSVT) sources, satellite imagery data, or data collected by a survey vehicle. It should be understood that the perception data includes images collected by perception sensors that are part of one or more vehicles, as well as semantic data. Although perception sensors are described, additionally or alternatively, the perception data is collected by infrastructure perception sensors (e.g., roadside cameras). Some examples of map data versions include, but are not limited to, crowdsourced map data, telemetry-based map data, and aerial map data. It should also be understood that high-definition map data represents ground-truth map data collected and verified by a survey vehicle.

[0048] In one non-limiting embodiment, the map data is created specifically for an automated driving system (e.g., an automated driving system (ADS) or an advanced driver assistance system (ADAS). In another embodiment, the map data is created for an inventory of road features, control devices, etc. In one embodiment, other dynamic map layers may be introduced on top of the map data, such as a map layer indicating the road surface or traffic conditions. As explained below, the system 10 fuses two or more versions of map data together to create fused map data 30 ( Figure 2 ). Since the fused map data 30 is based on more than one data source, the fused map data 30 includes higher precision and accuracy when compared to the original version of the map data. Figure 1 In the non-limiting embodiment shown, the one or more central computers 20 are in wireless communication with the one or more autonomous vehicles 26. After determining the fused map data 30, the one or more central computers 20 may share the fused map data 30 with the one or more autonomous vehicles 26.

[0049] Figure 2 It is shown Figure 1 A block diagram of the software architecture of one or more central computers 20 is shown. Figure 2In the example shown, the one or more central computers 20 include an error model 40, a measurement module 42, a cost function module 44, a smoothing module 46, and a state vector module 48. The error model 40 of the one or more central computers 20 receives as input two or more versions of map data points 70, 72 and ground truth map data 74. Specifically, in Figure 2 In the example shown, the error model 40 receives a first set of map data points 70 corresponding to a first version of the map data and a second set of map data points 72 corresponding to a second version of the map data. In one embodiment, the map data points 70, 72 represent a path along a road 92 ( Figure 3 It should be understood that the two groups of map data points 70, 72 are associated with each other and thus represent the same map attributes, such as lane lines, etc.

[0050] In one embodiment, the error model 40 may also receive ground truth map data 74 as input, however, it should be understood that the ground truth map data may not be available. It should also be understood that the high-definition map data represents ground truth map data. The first set of map data points 70 includes a first confidence value corresponding to the first version of the map data, and the second set of map data points 72 includes a second confidence value corresponding to the second version of the map data. The error model 40 of one or more central computers 20 executes one or more machine learning algorithms to determine the first deviation and first variance of the first set of map data points 70 and the second deviation and second variance of the second set of map data points 72. When the ground truth map data 74 is included, the error model 40 executes an iterative algorithm, such as a random sampling consensus (RANSAC) voting algorithm, to remove outliers from the first version of the map data, the second set of map data points 72, and the ground truth map data 74. In an embodiment, the ground truth map data 74 is based on an unbiased sampling of a road data set.

[0051] The measurement module 42 obtains data from one or more communication networks 24 ( Figure 1 ) receives a first set of map data points 70, a first deviation and a first variance, a second set of map data points 72, a second deviation and a second variance, ground truth map data 74 (if applicable), and a road network 88 representing a predefined geo-fenced area ( Figure 3 ) as input. One example of road network data 86 is Open Street Map (OSM). However, it should be understood that other types of road network data may also be used.

[0052] Figure 3 is an exemplary illustration of a portion of road network 88. Figure 2 and Figure 3, the road network 88 includes a discrete random curve 90, which represents lane lines arranged along a road 92. The discrete random curve 90 is defined by a series of discrete arcs 84, which are connected by a plurality of nodes 94 and include a plurality of state vectors s k , where k represents the state vector s k The total number of, k = 1, ..., N, N represents the state vector s k The discrete random curve 90 is a broken line curve or a Markov random curve, in which only the state vectors s that are continuously positioned relative to each other k are related to each other. A series of state vectors s k A step size Δs is defined, where the step size Δs represents two consecutive state vectors s located directly adjacent to each other along the discrete random curve 90 k The distance between them, and represents the spatial equivalent of the standard time series Δt of the Kalman filter.

[0053] Continue to refer to Figure 3 , each node 94 is located between two continuous state vectors s k The state vector s k is the state variable l k ,φ k Defined as a two-dimensional vector, where l k Denotes the corresponding state vector s k The position is expressed in Cartesian coordinates (x k ,y k ) indicates that φ k Denotes the corresponding state vector s k As described below, system 10 estimates each state vector s k Location i and the tangent angle φ i , where each state vector s k Location i and the tangent angle φ i Indicates the fused map data 30( Figure 2 ). Figure 3 As shown, the first set of map data points 70 and the second set of map data points 72 are also superimposed on the road network 88 .

[0054] The first set of map data points 70 is a portion of a first straight line curve M(s), wherein the first set of map data points 70 are respectively represented as a first series of points o m,k =M(s k ), k=1, ..., N, the second set of map data points 72 is a part of the second straight line curve H(s), wherein the second set of map data points 72 are respectively represented as a second series of points o h,k =H(sk ), k=0,…,N. Figure 4 yes Figure 3 An enlarged view of a portion of a discrete random curve 90 is shown, the portion starting from the state vector s k-2 and ends with the state vector s k+1 . Reference Figure 2 , Figure 3 and Figure 4 , random variable L k ( Figure 4 ) corresponds to each node 94 of the discrete random curve 90 and represents each state vector s k Location k ( Figure 2 The measurement module 42 determines the corresponding first map data point o m,i and the corresponding second map data point o h,i The symbol distance d i , where the symbol distance d i including the first map data point o from the corresponding m,k The first perpendicular distance measured to the discrete random curve 90 and from the corresponding second map data point o h,k , or i =(d m,k ,d h,k ) to the second perpendicular distance d of the discrete random curve 90 h，k .

[0055] Reference Figure 2 The cost function module 44 of one or more central computers 20 receives the symbol distance d from the measurement module 42 i , and by minimizing the spatial Kalman filter cost function based on the signed distance d i and the tangent angle φ k Estimate k state vectors s k Location k , the state vector s k Location k is a part of the discrete random curve 90 ( Figure 2 ). It should be understood that the spatial Kalman cost function represents the function that leads to the determination of the function for estimating the position l k It should be understood that because the discrete random curve 90 ( Figure 2 ) represents the spatial equivalent of the standard time series Δt, so the spatial Kalman filter technique is used to fuse two or more versions of map data together to create fused map data 30. The spatial Kalman filter cost function is expressed as Equation 1:

[0056]

[0057] in, Denotes the corresponding state vector s k The estimated position of v represents the deviation of measurement noise, φ i Denotes the corresponding state vector s k The tangent angle, q u represents the variance of additive white Gaussian noise, It should be understood that the measurement noise μ v The deviation of is determined based on the first deviation of the first set of map data points 70 and the second deviation of the second set of map data points 72. It should also be understood that if ground truth map data 74 is provided, the covariance of the measurement noise is determined based on the ground truth map data 74. Otherwise, the covariance of the measurement noise is determined based on crowdsourced map data received from multiple vehicles and / or vehicles passing by

[0058] It should be understood that the spatial Kalman filter cost function filters through the discrete random curve 90 ( Figure 3 )’s state vector s k The smoothing module 46 receives the state vector s from the cost function module 44 of one or more central computers 20. k Then, the smoothing module 46 recursively traverses the state vector s of the discrete random curve 90 k Execute the Kalman smoothing function, where the backward recursion includes the latest state vector s k The Kalman smoothing function is expressed as Equation 2:

[0059]

[0060] in, Denotes the corresponding state vector s k The estimated position of the Kalman smoothing function estimates the position of the N state vectors l k and the tangent angle φ k ( Figure 2 ), the position of the state vector l k and the tangent angle φ k is a portion of a discrete random curve 90, where the state vector s k Each represents a map point of the fused map data 30 .

[0061] It should be appreciated that the cost of executing the Kalman smoothing function described in Equation 2 increases with the number of k state vectors s k The size of the discrete random curve 90 ( Figure 3 ) represents the new state vector sk Therefore, the state vector module 48 responds to the introduction of a new state vector s into the discrete random curve 90. k To estimate the updated state vector The updated state vector Indicates the location of the new data point l k and the tangent angle φ k The updated state vector is expressed in Equation 3 as:

[0062]

[0063] in, represents the predicted state, d k represents a matrix indicating the lateral distance and the longitudinal distance of the first map data point and the lateral distance and the longitudinal distance of the second map data point, represents the covariance matrix.

[0064] Figure 5 is a process flow diagram illustrating a method 500 for fusing two or more versions of map data together. Figure 1-5 , the method 500 may begin at block 502. In block 502, the error model 40 of the one or more central computers 20 executes one or more machine learning algorithms to determine a first bias and a first variance of the first set of map data points 70 and a second bias and a second variance of the second set of map data points 72. In the case of including the ground truth map data 74, the error model 40 executes an iterative algorithm, such as a RANSAC voting algorithm, to remove outliers from the first version of the map data, the second set of map data points 72, and the ground truth map data 74. The method 500 may then proceed to block 504.

[0065] In block 504, the measurement module 42 obtains information from one or more communication networks 24 ( Figure 1 ) receives a first set of map data points 70, a first deviation and a first variance, a second set of map data points 72, a second deviation and a second variance, ground truth map data 74 (if applicable), and a road network 88 representing a predefined geo-fenced area ( Figure 3 ) as input. As described above, the road network 88 includes discrete random curves 90, which represent lane lines set along the road 92. Then, the method 500 can proceed to box 506.

[0066] In block 506, the measurement module 42 of one or more central computers 20 determines the symbol distance d i , where the symbol distance d i including the first map data point o from the correspondingm,k The first perpendicular distance d measured to the discrete random curve 90 m,K and from the corresponding second map data point o h,k , or i =(d m,k ,d h,k ) to the discrete random curve 90 (see Figure 3 ) of the second vertical distance d h,k Method 500 may then proceed to block 508 .

[0067] In block 508, the cost function module 44 of the one or more central computers 20 performs the spatial Kalman filter cost function based on the symbol distance d i and the tangent angle φ k To estimate k state vectors s as part of a discrete random curve 90 k Location k The method 500 may then proceed to block 510 .

[0068] Next, in block 510, the smoothing module 46 of one or more central computers 20 executes the Kalman smoothing function to estimate the N state vectors s k Location k and the tangent angle φ k ( Figure 2 ), state vector s k Location k and the tangent angle φ k is a portion of a discrete random curve 90, where the state vector s k Each represents a map point of the fused map data 30. Then, the method 500 may terminate.

[0069] Referring generally to the accompanying drawings, the disclosed system for fusing two different versions of map data together has various technical effects and benefits. Specifically, the fused map data provides greater integrity, precision, and accuracy than the original versions of the map data, and also includes enhanced robustness. It should be understood that the discrete random curves used to simulate lane markings or edge features along the road can accommodate any type of unique road geometry, such as clothoids, etc. It should also be understood that the current method of fusing map data together is scalable and can therefore be extended to multiple lanes and large-scale cloud-based maps.

[0070] The central computer may refer to, for example, an electronic circuit, a combinational logic circuit, a field programmable gate array (FPGA), a processor (shared, dedicated or group) that executes code, or a combination of some or all of the above, or a part of the above electronic circuit, combinational logic circuit, field programmable gate array (FPGA) or processor. In addition, the controller may be microprocessor-based, such as a computer having at least one processor, memory (RAM and / or ROM), and associated input and output buses. The processor may operate under the control of an operating system residing in the memory. The operating system may manage computer resources so that computer program code embodied as one or more computer software applications (e.g., applications residing in the memory) may have instructions executed by the processor. In an alternative embodiment, the processor may execute the application directly, in which case the operating system may be omitted.

[0071] The description of the present disclosure is merely exemplary in nature, and variations that do not depart from the gist of the present disclosure are intended to fall within the scope of the present disclosure. These variations should not be regarded as departing from the spirit and scope of the present disclosure.

Claims

1. A system for fusing two or more versions of map data together to create fused map data, the system comprising: one or more central computers storing the two or more versions of map data and ground truth map data, wherein each version of the map data represents a predefined geo-fenced area, wherein the one or more central computers execute instructions to: receiving road network data representing a road network of the predefined geo-fenced area, a first set of map data points, and a second set of map data points, wherein the road network data comprises a discrete random curve representing lane markings, wherein the discrete random curve comprises a plurality of state vectors, each state vector being defined by a respective position and a tangent angle; determining a signed distance, wherein the signed distance comprises a first perpendicular distance measured from a corresponding first map data point to the discrete random curve and a second perpendicular distance measured from a corresponding second map data point to the discrete random curve; estimating the position of the state vector of the discrete random curve based on the signed distance and the tangent angle by minimizing a spatial Kalman filter cost function; and A Kalman smoothing function is executed to estimate the position and tangent angle of the state vector, the position and tangent angle of the state vector being a part of the discrete random curve, wherein the state vectors respectively represent a map point of the fused map data.

2. The system according to claim 1, wherein: The discrete random curve is one of the following: a broken line curve and a Markov random curve.

3. The system according to claim 1, wherein: The one or more central computers execute instructions to: detecting a data point representing a new state vector introduced into the discrete random curve; and In response to detecting the data point, an updated state vector is estimated, wherein the updated state vector indicates the position of the new data point and a corresponding value of the tangent angle.

4. The system according to claim 3, wherein: The updated state vector is expressed as: in, represents the updated state vector, k represents the total number of state vectors, represents the predicted state, d k represents a matrix indicating the lateral distance and the longitudinal distance of the first map data point and the lateral distance and the longitudinal distance of the second map data point, μ v represents the deviation of the measurement noise, represents the covariance of the measurement noise, represents the covariance matrix.

5. The system according to claim 3, wherein: The spatial Kalman filter cost function is expressed as: in, Denotes the updated state vector s k The estimated position of k represents the total number of state vectors, d i represents the signed distance between the first map data point and the second map data point, μ v represents the deviation of the measurement noise, q u represents the variance of additive white Gaussian noise, represents the covariance of the measurement noise, φ i represents the tangent angle.

6. The system according to claim 5, wherein: The one or more central computers execute instructions to: receiving ground truth map data; executing one or more machine learning algorithms to determine a first bias and a first variance of the first set of map data points and a second bias and a second variance of the second set of map data points; as well as An iterative algorithm is performed to remove outliers from the first map data points, the second map data points, and the ground truth map data.

7. The system according to claim 6, wherein: The covariance of the measurement noise is determined based on the ground truth map data.

8. The system according to claim 6, wherein: The iterative algorithm is a random sample consensus (RANSAC) voting algorithm.

9. The system according to claim 8, wherein: The deviation of the measurement noise is determined based on the first deviation of the first set of map data points and the second deviation of the second set of map data points.

10. The system according to claim 3, wherein: The Kalman smoothing function is expressed as: in, Denotes the updated state vector s k The estimated position of N represents the total number of state vectors plus one, d i represents the signed distance between the first map data point and the second map data point, μ v represents the deviation of the measurement noise, q u represents the variance of additive white Gaussian noise, represents the covariance of the measurement noise, φ i represents the tangent angle.

Images