A geomagnetic road positioning method, system and storage medium combining coarse and fine matching

By using a coarse and fine matching method of a random forest classifier and a multi-scale twin neural network in geomagnetic road navigation, the problems of large calculation volume and few feature dimensions of the geomagnetic navigation algorithm are solved, the positioning accuracy and reliability are improved, and the noise impact is reduced.

CN118329014BInactive Publication Date: 2024-11-15HARBIN INST OF TECH AT WEIHAI
View PDF 4 Cites 0 Cited by

Patent Information

Application Number
CN202410488155.8
Authority / Receiving Office
CN · China
Patent Type
Patents(China)
Current Assignee / Owner
Filing Date
2024-04-23
Publication Date
2024-11-15
Estimated Expiration
Not applicable · inactive patent

AI Technical Summary

Technical Problem

The existing geomagnetic road navigation algorithm has a large amount of calculation and few geomagnetic data feature dimensions, which makes it difficult to improve positioning accuracy and reliability, and is susceptible to measurement noise.

Method used

The geomagnetic road positioning method combined with coarse and fine matching is adopted. By establishing a spatial sequence geomagnetic reference library, using a random forest classifier for coarse matching, and combining multi-scale twin neural network for fine matching, increasing the characteristic dimensions of the geomagnetic reference library and reducing the impact of measurement noise. , improve positioning accuracy.

Benefits of technology

It effectively improves the vehicle's road positioning accuracy based on geomagnetic information, reduces the calculation amount and the probability of mismatch, and improves the real-time and accuracy of positioning.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN118329014B_ABST
    Figure CN118329014B_ABST
Patent Text Reader

Abstract

The present invention provides a geomagnetic road positioning method, system and storage medium combining coarse and fine matching, which relates to the field of intelligent driving, and is to solve the problem that the existing method has a large amount of calculation, and the characteristic dimension of geomagnetic data is small, which makes the algorithm susceptible to measurement noise. It includes: step one, obtaining the geomagnetic scalar, three-axis geomagnetic vector, and three-axis attitude angle information of the vehicle along the path, and establishing a spatial sequence geomagnetic reference library; step two, sampling the reference library, obtaining a training set pair to train a pre-built random forest classifier, and obtain a coarse matching model; step three, obtaining a training set pair to train a pre-built multi-scale twin neural network, and obtain a fine matching model; step four, obtaining the spatial sequence of geomagnetic intensity and attitude angle when the vehicle is driving; step five, using the coarse matching model to determine the road where the current vehicle is located, and using the fine matching model to determine the position of the current vehicle in the road; step six, repeating steps four to five until the positioning is completed.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] The present invention relates to the field of intelligent driving technology, and in particular to a geomagnetic road positioning method, system and storage medium combining coarse and fine matching. Background Art

[0002] In recent years, research on intelligent driving technology has become increasingly in-depth. Vehicles are increasingly demanding systems for environmental perception, positioning, and control. Accurate positioning is a fundamental requirement for implementing various navigation functions. Excessive positioning deviations can lead to driving safety issues and even accidents. Commonly used vehicle positioning technologies include odometers, inertial navigation systems (INS), and global positioning systems (GPS). Methods like odometers and INS can cause errors to accumulate over time during long-distance positioning, making them unable to provide consistently high-precision positioning and requiring corrections in combination with other methods. GPS positioning has no cumulative errors but is susceptible to occlusion and signal interference. Improving vehicle positioning accuracy in satellite-denied environments is a major challenge.

[0003] Geomagnetic navigation, which relies on the Earth's magnetic field and the magnetic structures in the surrounding environment to obtain positioning information, is a widely used positioning and navigation method. It is low-cost, fully regional, and error-free. Combined with odometers, INS, and other methods, it can provide long-distance, high-precision positioning for vehicles in satellite-denied environments. However, traditional geomagnetic road navigation algorithms require traversing every road within the navigation area. When vehicles navigate complex road networks, the navigation area is numerous, and traversing all roads is computationally intensive, resulting in a high probability of mismatches. This reduces the real-time and accuracy of positioning and places high demands on hardware resources. Geomagnetic data typically uses geomagnetic scalars or triaxial vectors as features, resulting in a low dimensionality. Furthermore, in real-world road environments, geomagnetic data is subject to measurement noise, external environmental factors, and other influences. Positioning based solely on geomagnetic scalar or vector data is prone to mismatches, making it difficult to improve positioning accuracy and reliability. Summary of the Invention

[0004] The technical problems to be solved by the present invention are:

[0005] The existing geomagnetic road navigation algorithm has a large amount of computation and the geomagnetic data has few feature dimensions, which makes the geomagnetic road navigation algorithm susceptible to measurement noise and makes it difficult to improve positioning accuracy and reliability.

[0006] The present invention is to solve the above technical problems using the following technical solutions:

[0007] The present invention provides a geomagnetic road positioning method combining coarse and fine matching, comprising the following steps:

[0008] Step 1: Measure the roads in the navigation area, obtain the geomagnetic scalar, three-axis geomagnetic vector, and three-axis attitude angle information of the vehicle along the path, and establish a spatial sequence geomagnetic reference library;

[0009] Step 2: Sample the benchmark library to obtain training set 1, and use the training set to train a pre-built random forest classifier to obtain a coarse matching model;

[0010] Step 3: Sample the benchmark library to obtain training set 2, and use training set 2 to train the pre-built multi-scale twin neural network to obtain a precise matching model;

[0011] Step 4: while the vehicle is traveling along the road, a spatial sequence of geomagnetic intensity and attitude angle is obtained;

[0012] Step 5: Use the coarse matching model to classify the sequences obtained during driving and determine the road the vehicle is currently on; use the fine matching model to locate the sequences obtained during driving and determine the position of the road the vehicle is currently on;

[0013] Step 6: Repeat steps 4 to 5 until a stop command is received from the user.

[0014] Furthermore, step one includes the following process:

[0015] Measure the roads in the navigation area and measure the three-axis attitude angle and roll angle A θ,t , pitch angle A γ,t and yaw angle Geomagnetic X component M X,t , Y component M Y,t and the Z component M Z,t ;

[0016] Convert the measured geomagnetic components into geomagnetic components in the geographic coordinate system Right now:

[0017]

[0018]

[0019]

[0020]

[0021] Among them C x C y C z is the coordinate axis transformation matrix;

[0022] Establish a spatial sequence geomagnetic benchmark library, expressed as:

[0023]

[0024] Where R i Represents the geomagnetic reference library sequence of the i-th road, i∈[1, N1], N1 is the number of roads, represents the geomagnetic scalar along the i-th road, and They represent the spatial sequence of the geomagnetic X component, geomagnetic Y component, geomagnetic Z component, roll angle, pitch angle and yaw angle along the i-th road, respectively. The spatial distance interval of each point is s0, and the sequence length is L 1,i , T represents transpose.

[0025] Furthermore, the method for obtaining the training set 1 in step 2 is: sampling the mth road in the geomagnetic reference library using a sliding window with a length of L2 and a step size of s1, and obtaining N 2,m A sequence of length L2, with a sequence label of m; all N1 roads in the geomagnetic reference database are sampled to obtain the total There are N1 sequences in total.

[0026] Furthermore, the method for obtaining the second training set in step 3 is as follows: randomly sample the mth road in the geomagnetic reference library N3 times, using 3 windows for each sampling, with the length of window 1 being L2, and the obtained sequence is defined as the reference sequence; the length of window 2 is a random number in the range [L2-L rand ,L2+L rand ], L rand As a parameter, there are at least Lno points that do not overlap between window 2 and window 1, and the obtained sequence is defined as a dissimilar sequence; the length of window 3 is a random number, and the range is the same as that of window 2. The end point of window 3 is aligned with that of window 1, and the obtained sequence is defined as a similar sequence. Each point P on the similar sequence except the end point is i Perform random sampling, that is:

[0027] P i =P rand ,P rand ∈[min(P i ,P i+1 ),max(P i ,P i+1 )]

[0028] Gaussian noise with a mean of 0 and a standard deviation of σ is added to the similar sequence; the reference sequence and the similar sequence obtained in each sampling are defined as a group of positive samples; the reference sequence and the dissimilar sequence obtained in each sampling are defined as a group of negative samples.

[0029] Furthermore, the network framework of the multi-scale twin neural network in step 3 includes two identical feature extraction modules that share network parameters, N LThe feature extraction module consists of N fully connected layers and Sigmoid layers. C One-dimensional convolution modules are composed of N C The output of each convolution module is subtracted from the output of the convolution module of the same level on the opposite side and flattened as the feature extracted at this scale. All features are concatenated and input into the fully connected layer for processing to determine the similarity of the two input samples. Finally, the output is normalized to 0-1 through the Sigmoid layer. The one-dimensional convolution module consists of a one-dimensional convolution layer, a LeakyReLU layer, and a one-dimensional maximum pooling layer.

[0030] Furthermore, the training process of the multi-scale twin neural network in step 3 is as follows:

[0031] Input 1 is a reference sequence, and input 2 is a similar or dissimilar sequence. Two feature extraction modules extract features from the two input sequences respectively. The two extracted features are subtracted and flattened, and after passing through the fully connected layer and the Softmax layer in sequence, the similarity of the two sequences is output. The model parameters are updated using the defined loss function.

[0032] Furthermore, the loss function is a binary cross entropy loss function.

[0033] Furthermore, step five includes the following process: when the vehicle is traveling along the road, each time positioning is performed, first obtaining a geomagnetic intensity and attitude angle measurement sequence of length L2, using a coarse matching model to classify the sequence obtained during driving, and determining the number of the current road;

[0034] On the roads with corresponding numbers in the geomagnetic reference library, a window of length L2 is used for sliding window sampling, and the sequence of each sampling is defined as a reference sequence. The sampled reference sequence and the measurement sequence are input into the multi-scale twin neural network, and the similarity result between the measurement sequence and the reference sequence is output. After the sliding window sampling is completed, the current vehicle position is determined.

[0035] A geomagnetic road positioning system combining coarse and fine matching has a program module corresponding to the steps of any one of the above technical solutions, and executes the steps of the above geomagnetic road positioning method combining coarse and fine matching during operation.

[0036] A computer-readable storage medium stores a computer program, wherein the computer program is configured to implement the steps of the geomagnetic road positioning method combining coarse and fine matching described in any one of the above technical solutions when called by a processor.

[0037] Compared with the prior art, the present invention has the following beneficial effects:

[0038] The present invention provides a geomagnetic road positioning method combining coarse and fine matching. According to the characteristics of road positioning, geomagnetic scalar and three-axis geomagnetic vector information are used in a geomagnetic reference library, and three-axis attitude angle information is added to increase the characteristic dimension of the geomagnetic reference library.

[0039] This method uses a coarse matching model based on a random forest algorithm to roughly match data sequences, determine the current road, and reduce the computational complexity of subsequent matching. It then uses a fine matching model based on a multi-scale twin neural network to precisely match the data sequence, reducing the impact of measurement noise and improving positioning accuracy. This method can effectively improve the accuracy of a vehicle's road positioning based on geomagnetic information. BRIEF DESCRIPTION OF THE DRAWINGS

[0040] Figure 1 Flowchart of the geomagnetic road positioning method combining coarse and fine matching in an embodiment of the present invention;

[0041] Figure 2 Schematic diagram of geomagnetic scalar sequence in an embodiment of the present invention;

[0042] Figure 3 Schematic diagram of the geomagnetic vector sequence in an embodiment of the present invention;

[0043] Figure 4 Schematic diagram of vehicle posture in an embodiment of the present invention;

[0044] Figure 5 Schematic diagram of a posture sequence in an embodiment of the present invention;

[0045] Figure 6 A schematic diagram of obtaining a coarse matching training set in an embodiment of the present invention;

[0046] Figure 7 is a sequence comparison diagram in the embodiments of the present invention;

[0047] Figure 8 Schematic diagram of a multi-scale twin neural network in an embodiment of the present invention;

[0048] Figure 9 A schematic diagram of obtaining a precise matching training set in an embodiment of the present invention;

[0049] Figure 10 2 is a comparison diagram of the alignment results of the models in the embodiment of the present invention. DETAILED DESCRIPTION

[0050] In order to enable those skilled in the art to better understand the present invention, exemplary embodiments or examples of the present invention will be described below with reference to the accompanying drawings. Obviously, the described embodiments or examples are only some of the embodiments or examples of the present invention, and not all of them. Based on the embodiments or examples of the present invention, all other embodiments or examples obtained by those skilled in the art without creative work should fall within the scope of protection of the present invention.

[0051] In order to make the above-mentioned objects, features and advantages of the present invention more obvious and easy to understand, specific embodiments of the present invention are described in detail below with reference to the accompanying drawings.

[0052] Specific implementation plan 1: Figure 1 As shown, the present invention provides a geomagnetic road positioning method combining coarse and fine matching, comprising the following steps:

[0053] Step 1: Measure the roads in the navigation area, obtain the geomagnetic scalar, three-axis geomagnetic vector, and three-axis attitude angle information of the vehicle along the path, and establish a spatial sequence geomagnetic reference library;

[0054] Step 2: Sample the benchmark library to obtain training set 1, and use the training set to train a pre-built random forest classifier to obtain a coarse matching model;

[0055] Step 3: Sample the benchmark library to obtain training set 2, and use training set 2 to train the pre-built multi-scale twin neural network to obtain a precise matching model;

[0056] Step 4: while the vehicle is traveling along the road, a spatial sequence of geomagnetic intensity and attitude angle is obtained;

[0057] Step 5: Use the coarse matching model to classify the sequences obtained during driving and determine the road the vehicle is currently on; use the fine matching model to locate the sequences obtained during driving and determine the position of the road the vehicle is currently on;

[0058] Step 6: Repeat steps 4 to 5 until a stop command is received from the user.

[0059] In the geomagnetic road positioning method, the geomagnetic reference library is stored as a spatial sequence along the path. Traditional methods use geomagnetic scalars or geomagnetic three-axis vectors as features, such as Figure 2 、 Figure 3 However, in real-world road environments, geomagnetic data is subject to measurement noise, external environmental factors, and other factors, resulting in errors. Positioning based solely on geomagnetic scalar or vector data is prone to mismatching and poor positioning accuracy.

[0060] Since the posture of a vehicle is restricted by the road when it is traveling along the road, the posture of the vehicle is roughly the same when passing the same point on the road. Therefore, the present invention adds the three-axis posture angle information of the vehicle along the path to the geomagnetic reference library. The three-axis posture angle is as follows: Figure 4 As shown, assuming that the coordinate axes X0, Y0, and Z0 of the vehicle itself coincide with the geographic coordinate system X, Y, and Z in the initial state, the posture of the vehicle at the current moment can be regarded as rotating around the three axes X0, Y0, and Z0 by a certain angle θ, γ, The result of θ, γ, They are respectively called the roll, pitch and yaw angles of the vehicle. The three-axis attitude sequence is as follows Figure 5 The present invention combines the three-axis attitude data with the geomagnetic scalar and geomagnetic three-axis vector data, expands the feature dimension of the reference library to 7 dimensions, and improves the accuracy and reliability of matching.

[0061] In order to reduce the impact of the complexity of the road network on the amount of calculation and accuracy, the present invention proposes to use a random forest classifier to perform coarse matching on the data sequence to determine the current road, so that the algorithm does not need to process other roads in the area, which greatly reduces the amount of calculation and improves the matching accuracy.

[0062] In geomagnetic road positioning, in order to match the geomagnetic sequence collected during vehicle driving with the reference library, it is usually necessary to align the points in the time series with the points in the reference library according to the spatial position based on the mileage information provided by navigation equipment such as odometers and INS, and convert them into a spatial sequence. However, due to the influence of geomagnetic and mileage data measurement noise, there is a certain error between the converted sequence and the reference map, such as Figure 7 Traditional matching algorithms such as DTW are prone to mismatching, resulting in reduced positioning accuracy. This paper proposes using a multi-scale twin neural network to precisely match data sequences, reducing the impact of noise in the sequence on the matching results and improving positioning accuracy.

[0063] Specific implementation plan 2: Measure the roads in the navigation area and measure the three-axis attitude angle and roll angle A θ,t , pitch angle A γ,t and yaw angle Geomagnetic X component M X,t , Y component M Y,t and the Z component M Z,t ;

[0064] Convert the measured geomagnetic components into geomagnetic components in the geographic coordinate system Right now:

[0065]

[0066]

[0067]

[0068]

[0069] Among them C x C y C z is the coordinate axis transformation matrix;

[0070] Establish a spatial sequence geomagnetic benchmark library, expressed as:

[0071]

[0072] Where R i Represents the geomagnetic reference library sequence of the i-th road, i∈[1, N1], N1 is the number of roads, represents the geomagnetic scalar along the i-th road, and They represent the spatial sequence of the geomagnetic X component, geomagnetic Y component, geomagnetic Z component, roll angle, pitch angle and yaw angle along the i-th road, respectively. The spatial distance interval of each point is s0, and the sequence length is L 1,i , T represents transposition. The rest of this embodiment is the same as the first embodiment.

[0073] Specific implementation plan three: Figure 6 As shown, the method for obtaining the training set 1 in step 2 is: use a sliding window with a length of L2 to sample the mth road in the geomagnetic reference library, with a sliding window step of s1, and obtain N 2,m A sequence of length L2, with a sequence label of m; all N1 roads in the geomagnetic reference database are sampled to obtain the total There are N1 sequences in total.

[0074] Before positioning begins, the random forest classifier is initialized and a pre-trained model is loaded. During driving, each positioning operation requires pre-collecting a spatial sequence of sensor geomagnetic intensity and attitude angle measurements over a period of time. This sequence is converted into a 7-dimensional spatial sequence of length L2. This measurement sequence is then input into the pre-trained classifier to obtain the classification result O1, which represents the current road. This implementation is otherwise identical to Specific Implementation Plan 2.

[0075] Specific implementation plan four: Figure 9 As shown, the method for obtaining the training set 2 in step 3 is as follows: randomly sample the mth road in the geomagnetic reference library N3 times, using 3 windows for each sampling, with the length of window 1 being L2, and the obtained sequence is defined as the reference sequence; the length of window 2 is a random number in the range [L2-L rand ,L2+L rand ], L randAs a parameter, there are at least Lno points that do not overlap between window 2 and window 1, and the obtained sequence is defined as a dissimilar sequence; the length of window 3 is a random number, and the range is the same as that of window 2. The end point of window 3 is aligned with that of window 1, and the obtained sequence is defined as a similar sequence. Each point P on the similar sequence except the end point is i Perform random sampling, that is:

[0076] P i =P rand ,P rand ∈[min(P i ,P i+1 ),max(P i ,P i+1 )]

[0077] Gaussian noise with a mean of 0 and a standard deviation of σ is added to the similar sequences. The reference sequence and the similar sequence obtained in each sampling are defined as a group of positive samples. The reference sequence and the dissimilar sequence obtained in each sampling are defined as a group of negative samples. A total of N1×N3 groups of positive samples and N1×N3 groups of negative samples are obtained through sampling. The rest of this embodiment is the same as the third embodiment.

[0078] Specific implementation plan five: Figure 8 As shown in the figure, the network framework of the multi-scale twin neural network in step 3 includes two identical feature extraction modules that share network parameters, N L The feature extraction module consists of N fully connected layers and Sigmoid layers. C One-dimensional convolution modules are composed of N C The output of each convolution module is subtracted from the output of the convolution module of the same level on the opposite side and flattened as the features extracted at this scale. All features are spliced ​​and input into the fully connected layer for processing to determine the similarity of the two input samples. Finally, the output is normalized to 0-1 through the Sigmoid layer. The one-dimensional convolution module consists of a one-dimensional convolution layer, a LeakyReLU layer, and a one-dimensional maximum pooling layer. The rest of this embodiment is the same as the specific embodiment 1.

[0079] In this implementation, the multi-scale twin neural network can ensure that the multi-scale features extracted by each convolutional layer are retained, effectively capture the details and overall information in the geomagnetic and angle sequences, reduce the influence of factors such as measurement noise, and improve the robustness and generalization ability of the network.

[0080] Specific implementation plan six: The training process of the multi-scale twin neural network described in step three is:

[0081] Input 1 is a reference sequence, and input 2 is a similar or dissimilar sequence. Two feature extraction modules extract features from the two input sequences, respectively. The two extracted features are subtracted and flattened, and after passing through a fully connected layer and a softmax layer, the similarity between the two sequences is output. The model parameters are updated using a defined loss function. This implementation is otherwise identical to Specific Implementation Plan 5.

[0082] Specific embodiment 7: The loss function is a binary cross entropy loss function. The rest of this embodiment is the same as specific embodiment 6.

[0083] Specific implementation plan eight: Step five includes the following process: when the vehicle is traveling along the road, each time positioning is performed, first obtaining a geomagnetic intensity and attitude angle measurement sequence of length L2, using a coarse matching model to classify the sequence obtained during driving, and determining the number of the current road;

[0084] On the roads corresponding to the number in the geomagnetic reference database, a sliding window sampling is performed using a window of length L2, with a step size of s2. The sequence of each sampling is defined as the reference sequence. The sampled reference sequence and the measurement sequence are input into the multi-scale twin neural network, and the similarity between the measurement sequence and the reference sequence is output. After the sliding window sampling is completed, the current vehicle location is determined. This embodiment is otherwise identical to the first embodiment.

[0085] The geomagnetic road positioning method (algorithm) proposed in the present invention, which combines coarse and fine matching, is the underlying technical core of the present invention. Various products can be derived based on the algorithm.

[0086] Based on the method proposed in the present invention, a geomagnetic road positioning system combining coarse and fine matching is developed using a programming language. The system has a program module corresponding to the steps of the above-mentioned technical solution, and executes the steps in the above-mentioned geomagnetic road positioning method combining coarse and fine matching during operation.

[0087] The developed system (software) computer program is stored on a computer-readable storage medium. The computer program is configured to implement the steps of the above-mentioned coarse-fine matching combined geomagnetic road positioning method when called by a processor. This materializes the present invention on a carrier, becoming a computer program product.

[0088] Various implementations of the systems and techniques described herein can be realized in digital electronic circuitry, integrated circuitry, dedicated ASICs (application specific integrated circuits), computer hardware, firmware, software, and / or combinations thereof. These various implementations can include being implemented in one or more computer programs that are executable and / or interpreted on a programmable system comprising at least one programmable processor, which can be a special purpose or general purpose programmable processor that can receive data and instructions from a storage system, at least one input device, and at least one output device, and transmit data and instructions to the storage system, the at least one input device, and the at least one output device.

[0089] The computer programs (also referred to as programs, software, software applications, or code) of the present invention include machine instructions for a programmable processor and can be implemented using high-level procedural and / or object-oriented programming languages, and / or assembly / machine languages. As used herein, the terms "machine-readable medium" and "computer-readable medium" refer to any computer program product, device, and / or apparatus (e.g., a magnetic disk, an optical disk, a memory, a programmable logic device (PLD)) for providing machine instructions and / or data to a programmable processor, including a machine-readable medium that receives machine instructions as a machine-readable signal. The term "machine-readable signal" refers to any signal for providing machine instructions and / or data to a programmable processor.

[0090] Example 1

[0091] In order to verify the accuracy of the method of the present invention, the method of the present invention is described in detail using the following examples.

[0092] Experimental scenario: The positioning method of this invention was applied to a simulated road scenario and compared with the commonly used road positioning algorithm, DTW. The simulation environment was 1500m x 600m, and the maximum difference in a single component of geomagnetic intensity within the area was 10,000nT. A vehicle was assumed to travel along one of the six roads in the benchmark database. The vehicle's reference position and the corresponding simulated geomagnetic and attitude angle data were obtained. The vehicle was positioned using the simulated data and compared with the reference position.

[0093] In this embodiment, the number of roads N1 is 6, the spatial distance interval s0 of each point is 1m, the sequence length L1 of each road benchmark library is 500 to 1600, the sequence length L2 is 15, the step size s1 of the sliding window is 1, and the parameter L rand Take 2, parameter Lno takes 2, standard deviation σ takes 1000nT, and the number of one-dimensional convolution modules N C Take 3, the number of fully connected layers and Sigmoid layers is N L Take 3.

[0094] Experimental environment: CPU: 12th Gen Intel(R) Core(TM) i5-1240P 1.70GHz, 16G RAM, Windows 11.

[0095] Comparing the DTW positioning results, random forest + DTW positioning results, and random forest + multi-scale twin neural network positioning results with the reference position, the paths obtained are as follows Figure 10 The corresponding indicators are shown in Table 1.

[0096] Table 1

[0097]

[0098] Depend on Figure 10 As shown in Table 1, the results of matching using the traditional DTW algorithm are prone to mismatches, with many incorrect matching results and a large average positioning error. By adding a coarse matching method based on the random forest algorithm, the DTW algorithm's road matching accuracy improved and the average positioning error decreased. Using a multi-scale twin neural network instead of the DTW algorithm for fine matching further reduced the average positioning error. This demonstrates that the proposed algorithm outperforms traditional methods in terms of average error and road matching accuracy, effectively improving the accuracy of geomagnetic road positioning.

[0099] Although the present invention is disclosed as above, the scope of protection disclosed by the present invention is not limited thereto. Those skilled in the art of the present invention may make various changes and modifications without departing from the spirit and scope of the present invention, and these changes and modifications will fall within the scope of protection of the present invention.

Claims

1. A geomagnetic road positioning method combining coarse and fine matching, characterized in that: The steps include: Step 1: Measure the roads in the navigation area, obtain the geomagnetic scalar, three-axis geomagnetic vector, and three-axis attitude angle information of the vehicle along the path, and establish a spatial sequence geomagnetic reference library; Step 2: Sample the benchmark library to obtain training set 1, and use the training set 1 to train a pre-built random forest classifier to obtain a rough matching model; The method for obtaining the training set 1 is as follows: the mth road in the geomagnetic reference library is sampled using a sliding window with a length of L2 and a step length of s1, and N 2,m A sequence of length L2, with the label of the sequence being m; all N1 roads in the geomagnetic reference database are sampled to obtain the common There are N1 sequences with labels in total; Step 3: Sample the benchmark library to obtain training set 2, and use training set 2 to train the pre-built multi-scale twin neural network to obtain a precise matching model; The method for obtaining the second training set is as follows: randomly sampling the mth road in the geomagnetic reference library N3 times, using 3 windows for each sampling, the length of window 1 is L2, and the obtained sequence is defined as the reference sequence; the length of window 2 is a random number in the range of [L2-L rand ,L2+L rand ], L rand As a parameter, window 2 and window 1 have at least L no The points do not overlap, and the obtained sequence is defined as a dissimilar sequence; the length of window 3 is a random number, and the range is the same as that of window 2. The end point of window 3 is aligned with window 1, and the obtained sequence is defined as a similar sequence. Each point P except the end point on the similar sequence i Perform random sampling, that is: P i =P rand ,P rand ∈[min(P i ,P i+1 ),max(P i ,P i+1 )] Add Gaussian noise with mean 0 and standard deviation σ to similar sequences; The reference sequence and similar sequence obtained from each sampling are defined as a group of positive samples; the reference sequence and dissimilar sequence obtained from each sampling are defined as a group of negative samples; Step 4: When the vehicle is traveling along the road, a spatial sequence of geomagnetic intensity and attitude angle is obtained; Step 5: Use the coarse matching model to classify the sequences obtained during driving and determine the road where the current vehicle is located; Use the precise matching model to locate the sequence obtained during driving and determine the current position of the vehicle on the road; Step 6: Repeat steps 4 to 5 until a stop command is received from the user.

2. The geomagnetic road positioning method combining coarse and fine matching according to claim 1 is characterized in that: Step 1 includes the following process: Measure the roads in the navigation area and measure the three-axis attitude angle and roll angle A θ,t 、Pitch angle A γ,t and yaw angle Geomagnetic X component M X,t , Y component M Y,t and the Z component M Z,t ; Convert the measured geomagnetic components into geomagnetic components in the geographic coordinate system Right now: Among them C x C y C z is the coordinate axis transformation matrix; Establish a spatial sequence geomagnetic benchmark library, expressed as: Where R i Represents the geomagnetic reference library sequence of the i-th road, i∈[1,N1], N1 is the number of roads, represents the geomagnetic scalar along the ith road, and They represent the spatial sequence of the geomagnetic X component, geomagnetic Y component, geomagnetic Z component, roll angle, pitch angle and yaw angle along the i-th road, respectively. The spatial distance interval of each point is s0, and the sequence length is L 1,i , Indicates transpose.

3. The geomagnetic road positioning method combining coarse and fine matching according to claim 1, characterized in that: The network framework of the multi-scale twin neural network in step 3 includes two feature extraction modules that are completely identical and share network parameters, N L The feature extraction module consists of N fully connected layers and Sigmoid layers. C One-dimensional convolution modules, N C The output of a convolution module is subtracted from the output of the convolution module of the same level on the opposite side, and flattened as the feature extracted at this scale. All features are spliced ​​and input into the fully connected layer for processing to determine the similarity of the two input samples. Finally, the output is normalized to 0-1 through the Sigmoid layer. The one-dimensional convolution module consists of a one-dimensional convolution layer, a LeakyReLU layer and a one-dimensional maximum pooling layer.

4. The geomagnetic road positioning method combining coarse and fine matching according to claim 3 is characterized in that: The training process of the multi-scale twin neural network described in step 3 is: Input 1 is the reference sequence, and input 2 is a similar or dissimilar sequence; the two feature extraction modules extract the features of the two input sequences respectively; the two extracted features are subtracted and flattened, and after passing through the fully connected layer and the Softmax layer in turn, the similarity of the two sequences is output; the model parameters are updated through the defined loss function.

5. The geomagnetic road positioning method combining coarse and fine matching according to claim 4 is characterized in that: The loss function is a binary cross entropy loss function.

6. The geomagnetic road positioning method combining coarse and fine matching according to claim 1, characterized in that: Step 5 includes the following process: when the vehicle is traveling along the road, each time positioning is performed, firstly, a geomagnetic intensity and attitude angle measurement sequence of length L2 is obtained, and the sequence obtained during driving is classified using a coarse matching model to determine the number of the current road; On the roads with corresponding numbers in the geomagnetic reference library, a window with a length of L2 is used for sliding window sampling, and the sequence of each sampling is defined as a reference sequence; the sampled reference sequence and the measurement sequence are input into the multi-scale twin neural network, and the similarity result between the measurement sequence and the reference sequence is output; when the sliding window sampling is completed, the current vehicle position is determined.

7. A geomagnetic road positioning system combining coarse and fine matching, characterized in that: The system has a program module corresponding to the steps of any one of claims 1 to 6 above, and executes the steps in the above-mentioned geomagnetic road positioning method combining coarse and fine matching when running.

8. A computer-readable storage medium, characterized in that: The computer-readable storage medium stores a computer program, and the computer program is configured to implement the steps of the geomagnetic road positioning method combining coarse and fine matching as described in any one of claims 1 to 6 when called by a processor.

Citation Information

Patent Citations

  • Mileage information assisted geomagnetic matching based vehicle positioning system and method

    CN107907134A

  • Geomagnetic reference library construction method for indoor vehicle geomagnetic matching positioning

    CN112504265A

  • Unmanned motion planning method based on digital twin

    CN113359709A

  • Passive combined indoor positioning system and method based on intelligent terminal sensor

    US20230304802A1