An automatic driving car transition scene positioning method based on multi-domain fusion of space-ground integration

Abstract

The application discloses a kind of heaven and earth integrated multi-domain fusion automatic driving car transition scene positioning method, obtains image in driving process, is matched with visual image feature to determine vehicle reaches transition scene area;Construct the combined positioning model based on robust adaptive, the combined positioning model includes GNSS / LEO / IMU combined positioning model, vehicle positioning is carried out to outdoor area;And LiDAR / IMU combined positioning model, vehicle positioning is carried out to indoor area;Interactive multi-model algorithm is used, and outdoor and indoor different positioning modes are smoothly switched in transition scene area.The application proposes an innovative solution to the defects of traditional positioning methods in switching scenes, which can effectively overcome the problem of poor positioning accuracy in current switching scenes, ensuring the continuity and stability of the trajectory of the automatic driving car during driving, and has huge market potential.

Description

Technical Field

[0001] This invention belongs to the field of autonomous vehicle positioning technology, specifically relating to a ground-air-ground multi-domain fusion method for positioning autonomous vehicles in transitional scenarios. Background Technology

[0002] Rapid technological advancements have made autonomous vehicles a crucial element of modern transportation. They not only enhance driving safety but also optimize traffic flow and reduce environmental pollution. Location technology is a core component of autonomous vehicles and a key to building modern intelligent transportation systems, supporting autonomous driving, route planning, and vehicle monitoring. This technology has evolved from initial single-technology applications to multi-technology integration and is now moving towards intelligent applications. With continuous technological upgrades, autonomous vehicle location technology will play an increasingly critical role in autonomous driving, traffic management, and smart city development.

[0003] Precise positioning is crucial for autonomous driving path planning, as positioning errors directly interfere with the accuracy of planning and control algorithms. Currently, autonomous vehicle positioning technology has made significant progress. Global Navigation Satellite Systems (GNSS) are the mainstream vehicle positioning method, due to their simple principle, low cost, and the fact that most devices use single-point GNSS positioning with accuracy typically at the meter level. In recent years, with the rise of Precise Point Positioning (PPP) and Real-Time Dynamic Differential (RTK) technologies, positioning accuracy has improved to the centimeter level. However, the canyon effect limits the use of GNSS alone. To compensate for the shortcomings of single technologies, combined positioning technologies based on GNSS and Inertial Navigation Systems (INS) have emerged in outdoor scenarios, significantly improving positioning accuracy and reliability through data fusion. Simultaneously, with the development of sensor technology, LiDAR, cameras, and ultrasonic sensors are widely used in vehicle positioning systems, enhancing the vehicle's perception of its surroundings and further improving positioning accuracy and stability. In addition to the vehicle's own positioning technology and strategies, these technologies have demonstrated excellent performance in both indoor and outdoor scenarios.

[0004] However, research on positioning technology in transitional indoor-outdoor scenarios is relatively scarce. GNSS signals are often interfered with before moving from open outdoor areas to indoor locations, tunnels, underground parking lots, under overpasses, etc., which greatly limits the positioning accuracy in these transitional scenarios and cannot meet the high-precision positioning requirements of autonomous vehicles. Summary of the Invention

[0005] To address the shortcomings of existing technologies, this application proposes a ground-based, multi-domain integrated positioning method for autonomous vehicles in transitional scenarios. This method utilizes LEO satellites, which offer numerous advantages such as low orbital speed, high power, strong penetration, relatively concentrated positioning error distribution, and stable positioning accuracy. Furthermore, it demonstrates superior robustness in complex terrain environments (mountains, canyons, forests, etc.) without relying on ground infrastructure. This invention employs a multi-sensor, multi-domain fusion, and comprehensive three-dimensional collaborative positioning mode, successfully achieving high-precision positioning of autonomous vehicles in transitional scenarios and seamless integration of indoor and outdoor positioning methods. This addresses the problem in existing autonomous vehicle positioning technologies where GNSS signals are susceptible to environmental limitations, leading to decreased positioning accuracy and interrupted positioning information.

[0006] The technical solution adopted in this invention is as follows:

[0007] A method for localization of autonomous vehicles in transitional scenarios through a multi-domain fusion approach integrating air, ground, and space includes the following steps:

[0008] Step 1: Acquire images during the driving process and determine the transition scene area of ​​the vehicle through visual image feature matching;

[0009] Step 2: Construct a robust and adaptive integrated positioning model, which includes a GNSS / LEO / IMU integrated positioning model for vehicle positioning in outdoor areas and a LiDAR / IMU integrated positioning model for vehicle positioning in indoor areas.

[0010] Step 3: Use an interactive multi-model algorithm to smoothly switch between different positioning methods for outdoor and indoor environments in transitional scene areas.

[0011] Furthermore, the method for determining transition scenarios is as follows:

[0012] Step 1-1: Acquire images of the autonomous vehicle during its driving process;

[0013] Steps 1-2: Divide the image into multiple sub-image blocks, calculate the number of main color pixels in each sub-image, extract the main color and its pixel count in the sub-image blocks, and construct the main color feature vector of the entire image.

[0014] Steps 1-3: Take photos of the road scene by setting data sampling points for the transition scene, and assemble the sampling point images into a transition scene database according to the main color feature vector descriptor;

[0015] Steps 1-4: During the driving process, the autonomous vehicle will match the collected image information with the transition scene image database to determine whether the vehicle has reached the transition scene area.

[0016] Furthermore, in steps 1-2, the K-means clustering algorithm is used to calculate the number of main color pixels in each sub-image block. Specifically, all pixels in the sub-image are regarded as data points, and the RGB value of each pixel is used as its feature vector.

[0017] Furthermore, in steps 1-4, the one-dimensional dynamic programming matching result of the two principal color feature vectors in each group is calculated iteratively. Let Γ au and Λ bv These are the main color feature vectors of the two images. Dynamic programming matching technology is used to determine the matching distance between the image to be matched and the images in the transition scene database. When the matching distance is greater than the set threshold ζ, it is determined that the vehicle has reached the transition scene area.

[0018] Furthermore, the GNSS / LEO / IMU integrated positioning model is constructed based on GNSS, LEO, and IMU, using the positioning information vector x output by the GNSS positioning module or the LEO positioning module. G and the positioning information vector x output by the IMU positioning module. I Location information is fused, and represented as: The fused positioning information is then processed using a robust filtering algorithm.

[0019] Furthermore, the switching between the GNSS positioning module and the LEO positioning module is based on the following: when the number of stable visible satellites is greater than or equal to 4, the positioning information vector x output by the GNSS positioning module is used. G When the number of stable visible satellites is less than 4, the positioning information vector x output by LEO is used. G .

[0020] Furthermore, a LiDAR / IMU integrated positioning model is constructed based on LiDAR and IMU, and the positioning information vector x is obtained based on LiDAR. L and the positioning information vector x obtained based on IMU I Location information is fused, and represented as: The fused positioning information is then processed using a robust filtering algorithm.

[0021] Furthermore, the following scheme for constructing seamless indoor and outdoor positioning using an interactive multi-model algorithm is proposed:

[0022] Step 3-1: Combine the outdoor positioning results obtained from the GNSS / LEO / IMU integrated positioning model and the indoor positioning results obtained from the LiDAR / IMU integrated positioning model. i The states of the hybrid system are obtained by interacting with (k-1) and the transition probability matrix Π. and variance q i (k-1);

[0023] Step 3-2: Use robust filtering algorithm to estimate and update system state variables and state covariance, and incorporate the results into the final output interaction.

[0024] Step 3-3: In the transitional scenario, update the model probability based on the likelihood function of the indoor-outdoor combined positioning model;

[0025] Steps 3-4: Based on the updated model probabilities, weight and fuse the robust filtering results of each sub-model in the hybrid model set, output the optimal estimated state, and determine the transition scene positioning information.

[0026] Furthermore, dynamic programming matching technology determines the matching distance between the two as follows:

[0027]

[0028] Among them, Γ au Λ represents the principal color feature vector of the u-th currently acquired image. bv Let d(Γ) represent the principal color feature vector of the image in the image database for the v-th transition scene. au ,Λ bv ) represents Γ au and Λ bv The element distance between them, Γ[a,u] represents Γ au The value of Λ[b,v] represents Λ bv The values ​​are u = 1, 2, 3, ..., z, v = 1, 2, 3, ..., z, where z is the number of images.

[0029] The beneficial effects of this invention are:

[0030] (1) In view of the problem that traditional positioning methods face the problem of decreased positioning accuracy and interruption of positioning information in transitional scenarios, this invention proposes a high-level autonomous vehicle transitional scenario positioning method that is different from the current mainstream multi-sensor fusion, vehicle-road cooperative positioning and other methods. It is suitable for transitional scenarios and effectively solves the problem of positioning performance degradation in transitional scenarios, thus improving the positioning accuracy in the area.

[0031] (2) This invention innovatively transitions the outdoor GNSS / LEO / IMU combined positioning method to the indoor lidar / IMU combined positioning method in the transition scene area, realizing the positioning of autonomous vehicles in all scenarios.

[0032] (3) The present invention adopts a diversified integrated positioning method of multi-sensor fusion and multi-domain fusion of air and ground to achieve higher precision positioning and seamless switching in transition scenarios, which can be commercially applied on a large scale. Compared with traditional positioning technology, it effectively improves the continuity and smoothness of the trajectory of autonomous vehicles during driving and has broad market prospects. Attached Figure Description

[0033] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on the provided drawings without creative effort.

[0034] Figure 1 This is a flowchart illustrating the technical process of the present invention.

[0035] Figure 2 Interactive multi-model schematic diagram for indoor and outdoor combined positioning. Detailed Implementation

[0036] To make the objectives, technical solutions, and advantages of this invention clearer, the invention will be further described in detail below with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are for illustrative purposes only and are not intended to limit the invention.

[0037] This embodiment completes the positioning experiment of the present invention at the entrance of an underground parking garage. The technical flowchart of the present invention, which provides a high-level autonomous vehicle transition scene positioning method integrating space, air, and ground multi-domain fusion, is shown below. Figure 1 As shown, the specific implementation steps are as follows:

[0038] Step 1: Determine the transition scene area of ​​the vehicle by matching visual image features.

[0039] Step 1-1: Use the vehicle-mounted image acquisition device to acquire images of the autonomous vehicle during its driving process;

[0040] Steps 1-2: Divide the acquired current frame image into m×n sub-image blocks. In this embodiment, m=n=5. Use the K-means clustering algorithm to calculate the number of dominant color pixels in each sub-image block. Specifically, treat all pixels in the sub-image as data points, and use the RGB value of each pixel as its feature vector. Select an appropriate number of cluster centers K, K=15, and run the K-means clustering algorithm until convergence. Finally, each cluster center represents a dominant color, and the number of pixels belonging to that cluster center is the number of pixels of that dominant color. In this way, the dominant color and its pixel count in the sub-image blocks are accurately extracted. Then, the frequency of each dominant color in each sub-block is integrated to construct the dominant color feature vector of the entire image. The color content of the current frame image can then be represented by a dominant color feature vector of size m×n×K=375.

[0041] Steps 1-3: In this invention, data sampling points are set for the transition scene to take pictures of the road scene, and the sampling point images are combined into a transition scene database according to the main color feature vector descriptor.

[0042] Steps 1-4: During the driving process, the autonomous vehicle matches the collected image information with the transition scene image database. Specifically, it uses a one-dimensional dynamic programming (DP) method to iteratively calculate the matching results of each pair of principal color feature vectors. Let Γ be the matching result. au and Λ bv Given the principal color feature vectors of two images, the matching distance between them is determined using dynamic programming (DP) matching techniques:

[0043]

[0044] Among them, Γ au Λ represents the principal color feature vector of the u-th currently acquired image. bv Let d(Γ) represent the principal color feature vector of the image in the image database for the v-th transition scene. au ,Λ bv ) represents Γ au and Λ bv The element distance between them, Γ[a,u] represents Γ au The value of Λ[b,v] represents Λ bv The values ​​are u = 1, 2, 3, ..., z, v = 1, 2, 3, ..., z, where z is the number of images.

[0045] Equation (1) can cover Γ a and Λ b The matrix represents the distances between all potential elements, and d(Γ) is placed within the matrix. au ,Λ bv The expression is simplified to d(u,v), and is described in detail below:

[0046]

[0047] To further calculate Γ a and Λ b The nearest feature vector matching distance between them, let the weights from d(u-1,v), d(u,v-1), and d(u-1,v-1) to d(u,v) be ω1, ω2, and ω3, respectively, where, ω3=τ·d(u,v), where τ is an empirical coefficient, specifically set as follows:

[0048]

[0049] Within the interval uv∈[-R,R], each of the above d(u,v) is connected according to its corresponding weight. A directed graph D is defined, with its initial point d(1,1) and its destination d(n,n). The shortest path length d between d(1,1) and d(n,n) is solved using Dijkstra's algorithm. min The feature vector matching distance between the current image to be matched and the images in the transition scene database is:

[0050]

[0051] Then, the image to be matched is compared with the images in the transition scene database in turn. When the matching distance is greater than the set threshold ζ, it is determined that the vehicle has reached the transition scene area.

[0052] Step 2: Construct a robust adaptive combined localization model.

[0053] Step 2-1: Construct an outdoor GNSS / LEO / IMU combined positioning model based on GNSS, LEO, and IMU respectively, to obtain accurate vehicle positioning information in outdoor areas that effectively utilize satellite signals.

[0054] The working principles of GNSS, LEO, and IMU are introduced below:

[0055] (1) GNSS positioning module: The location information received by the vehicle's GNSS signal receiver is: [x, y, z] T The coordinates of the i-th observable satellite are [x... (i) ,y (i ),z (i) ] T The actual distance between the receiver installed on the vehicle and the satellite is:

[0056]

[0057] Suppose there are n visible satellites that can provide positioning information, then a set of n pseudorange observation equations for visible satellites can be established:

[0058]

[0059] Where, ρ (i) Let δt be the pseudorange of the receiver relative to the i-th satellite, c represent the signal propagation speed, and δt be the pseudorange. u This refers to the receiver clock bias.

[0060] As can be seen from formula (4), there are 4 unknowns in the system of equations. Therefore, when the number of stable visible satellites is greater than or equal to 4, the system of equations can be solved to obtain GNSS positioning information. At the same time, RTK (Real-Time Kinematic) positioning technology can be used to correct the GNSS positioning information and output a higher precision positioning information vector x. G .

[0061] (2) LEO positioning module: If the number of stable visible satellites is less than 4, the GNSS satellite signal quality is considered unsatisfactory. In this case, LEO is needed for outdoor positioning, and the positioning information vector x is output. G .

[0062] (3) IMU Positioning Module: Utilizing the vehicle's acceleration a and angular velocity ω output by the Inertial Measurement Unit (IMU), the discrete-time data is integrated to obtain the vehicle's velocity v and position l in the world coordinate system. The calculation formula is as follows:

[0063]

[0064] Where v0 and l0 are the initial velocity and position information of the vehicle, a(t) is the acceleration a of the vehicle at time t, and v(t) is the velocity v of the vehicle at time t.

[0065] Furthermore, by integrating the angular velocity, the vehicle's heading angle can be obtained. The calculation formula is as follows:

[0066]

[0067] in, Let ω be the initial heading angle of the vehicle, and ω(t) be the angular velocity of the vehicle at time t.

[0068] Finally, by combining the vehicle's position information l and heading angle information, the vehicle's positioning information vector x is calculated. I .

[0069] (4) Based on the above GNSS, LEO, and IMU positioning modules, an outdoor scene GNSS / LEO / IMU combined positioning model is constructed. The specific process of fusing the positioning information of GNSS, LEO, and IMU is as follows:

[0070] Based on the positioning information vector x output by the aforementioned GNSS positioning module or LEO positioning module G and the positioning information vector x output by the IMU positioning module. I The location information is then fused, as shown in the following formula:

[0071]

[0072] Based on the nonlinear characteristics of the GNSS / LEO / IMU integrated positioning model, a robust filtering algorithm is adopted as the filtering algorithm for integrated positioning.

[0073] Integrated positioning systems can be described as follows:

[0074]

[0075] Where, x k For the combined positioning system parameter state vector, Φ k-1 Let T be the state transition matrix, and τ be the noise allocation matrix. k-1 Let z be the process noise vector. k H represents the observation vector of the integrated positioning system. k System observation matrix, v k This is the observed noise vector.

[0076] Robust estimation first obtains the state error and observation error equations of the integrated positioning system, as shown in the following equation:

[0077]

[0078] Among them, z k For the observation vectors of the combined positioning system, Let k be the state prediction vector at time k. This is the state estimate at time k.

[0079] In robust estimation, robust M-estimation is used on the observation vector to construct the following conditional extrema:

[0080]

[0081] in, For the observation vector z k The equivalent weight matrix, For the prediction vector The equivalent weight matrix.

[0082] This invention uses the IGGⅢ weight function for solving, and its expression is as follows:

[0083]

[0084] Therefore, the equivalent weight function is:

[0085]

[0086] In the formula, the values ​​of c0 and c1 are generally [1.5, 2.0] and [3.0, 8.5], respectively. Assign residual values ​​to its standard. σi For v i The mean squared error, b i The weight is the weight of the i-th observation.

[0087] Equation (11) for x k Finding the extrema yields the robust solution vector of the system's state parameters:

[0088]

[0089] The corresponding posterior covariance matrix can then be described as:

[0090]

[0091] The filter gain is:

[0092]

[0093] further, The recursive solution is:

[0094]

[0095] The posterior covariance matrix corresponding to the recursive solution can be approximately represented as:

[0096]

[0097] In the formula, express Antivariance equivalent covariance It means that z k The robust equivalent covariance, and

[0098] Therefore, the observed data can be classified by optimizing appropriate control parameters c0 and c1. When outliers appear in the observed data, the equivalent weight function will assign corresponding weights to them according to the size of the standard residuals, thereby reducing the interference of outlier observations on state estimation, exerting robustness, and further enhancing the stability of the filtered estimation.

[0099] By following the steps above, the fusion of GNSS, LEO, and IMU positioning information can be achieved, enabling more accurate vehicle positioning information to be obtained by effectively utilizing satellite signal areas outdoors.

[0100] Step 2-2: Based on LiDAR and IMU, construct an indoor LiDAR / IMU combined positioning model to achieve high-precision positioning of autonomous vehicles in areas without satellite signals (transitional scenarios).

[0101] Specifically:

[0102] (1) LiDAR:

[0103] First, point cloud data obtained from LiDAR scanning is acquired. This data is then processed to obtain the currently observed feature M. Based on the vehicle's pose and the pose of feature M in the vehicle coordinate system, the pose of feature M in the world coordinate system is calculated. Feature M is then added to the map (map update). When the vehicle's pose changes and feature M is observed again, the vehicle's pose x can be calculated based on the poses of feature M in both the world and vehicle coordinate systems. L .

[0104] The positioning information vector x obtained based on LiDAR L and the positioning information vector x obtained based on IMU I The splicing is performed as shown in the following formula:

[0105]

[0106] Taking into account the nonlinear characteristics of LiDAR / IMU combined positioning, a robust filtering algorithm is also used as the filtering algorithm for combined positioning. The specific steps are as described in step 2-2.

[0107] Step 3: Smoothly switch between different positioning methods for outdoor and indoor environments in the transitional scene area, as follows:

[0108] An interactive multi-model algorithm (IMM) is employed to construct a seamless indoor-outdoor positioning solution, achieving automatic and smooth switching between indoor and outdoor positioning modes and continuously outputting high-precision positioning results, such as... Figure 2 As shown, it specifically includes four parts: input information interaction, model filtering, model probability update, and estimation fusion.

[0109] Step 3-1: Input Information Interaction. This involves exchanging the outdoor positioning results obtained from the GNSS / LEO / IMU integrated positioning model in Step 2 with the indoor positioning results obtained from the LiDAR / IMU integrated positioning model. i The states of the hybrid system are obtained by interacting with (k-1) and the transition probability matrix Π. and variance q i (k-1), specifically described as:

[0110]

[0111] in, and q j (k-1) represents the state and variance of any model in the hybrid system at time k-1, μ j|i (k-1) is the mixing probability of switching any other model j to model i at time k-1, Π=[π ij ] M×MM is the number of sub-models, π ij μ is the probability of jumping from model i to model j. j (k-1) represents the probability of model j at time k-1, where i,j = 1, 2.

[0112] Step 3-2: Model Filtering Process. The outdoor and indoor combined positioning systems use robust filtering algorithms to estimate and update the system state variables and state covariance. The results from each filter are incorporated into the final output interaction.

[0113] The specific process is as follows:

[0114] For outdoor positioning, the results obtained from the interaction As the input to the robust filtering algorithm of the GNSS / LEO / IMU integrated positioning model, the robust filtering calculation is performed again as shown in Equation 8-18, and the output is...

[0115] For indoor positioning, the results obtained from the interaction As input to the robust filtering algorithm of the LiDAR / IMU integrated positioning model, robust filtering prediction calculation is performed again as shown in Equation 8-18, and the output is...

[0116] Step 3-3: Model Probability Update. In the transitional scenario, the process of updating the model probability based on the likelihood function of the indoor / outdoor combined positioning model is as follows:

[0117]

[0118]

[0119] Where, μ i (k) represents the probability of any model i in the mixture model at time k, Ξ i (k) is the likelihood function of model i at time k, v i S is the residual of model i at time k. i (k) is the residual covariance matrix of model i at time k.

[0120] Steps 3-4: Estimate the fusion. Based on the updated model probability μ i (k) Robust filtering results of each sub-model in the weighted fusion hybrid model set Output the optimal estimated state to determine the location information for the transition scene.

[0121]

[0122] This enables the transition from outdoor to indoor scene positioning. The switching from indoor to outdoor scene positioning also includes: identification of the transition scene, construction of a combined positioning model for the transition scene, and smooth switching of the transition scene positioning method. For details, refer to the process of achieving the transition from outdoor to indoor scene positioning. This completes the entire process of the integrated space-ground multi-domain fusion high-level autonomous vehicle transition scene positioning method designed in this invention.

[0123] In summary, the ground-air-ground multi-domain fusion high-level autonomous vehicle transition scenario positioning method designed in this invention solves the problems of decreased positioning accuracy and interruption of positioning signals in transition scenarios, and achieves the goal of high-precision positioning and smooth switching of autonomous vehicles in transition scenarios.

[0124] The above embodiments are only used to illustrate the design concept and features of the present invention, and their purpose is to enable those skilled in the art to understand the content of the present invention and implement it accordingly. The protection scope of the present invention is not limited to the above embodiments. Therefore, all equivalent changes or modifications made based on the principles and design ideas disclosed in the present invention are within the protection scope of the present invention.

Claims

1. A method for localization of transitional scenarios for autonomous vehicles integrating space, air, and ground domains, characterized in that, Includes the following steps: Step 1: Acquire images during the driving process and determine the transition scene area of ​​the vehicle through visual image feature matching; Step 2: Construct a robust and adaptive integrated positioning model, which includes a GNSS / LEO / IMU integrated positioning model for vehicle positioning in outdoor areas and a LiDAR / IMU integrated positioning model for vehicle positioning in indoor areas. Step 3: Employ an interactive multi-model algorithm to smoothly switch between different positioning methods for outdoor and indoor environments in transitional scene areas; the solution for seamless indoor and outdoor positioning using the interactive multi-model algorithm is as follows: Step 3-1: Combine the outdoor positioning results obtained from the GNSS / LEO / IMU integrated positioning model with the indoor positioning results obtained from the LiDAR / IMU integrated positioning model. and transition probability matrix Interact to obtain the state of the hybrid system. and variance ; Step 3-2: Use the robust filtering algorithm to estimate and update the system state variables and state covariance, and incorporate the results into the final output interaction. Step 3-3: In the transitional scenario, update the model probability based on the likelihood function of the indoor-outdoor combined positioning model; Steps 3-4: Based on the updated model probabilities, weight and fuse the robust filtering results of each sub-model in the hybrid model set, output the optimal estimated state, and determine the transition scene positioning information.

2. The method for positioning autonomous vehicles in transitional scenarios using a multi-domain integrated space-ground fusion system according to claim 1, characterized in that, The method for determining transition scenarios is as follows: Step 1-1: Acquire images of the autonomous vehicle during its driving process; Steps 1-2: Divide the image into multiple sub-image blocks, calculate the number of main color pixels in each sub-image, extract the main color and its pixel count in the sub-image blocks, and construct the main color feature vector of the entire image. Steps 1-3: Take photos of the road scene by setting data sampling points for the transition scene, and assemble the sampling point images into a transition scene database according to the main color feature vector descriptor; Steps 1-4: During the driving process, the autonomous vehicle will match the collected image information with the transition scene image database to determine whether the vehicle has reached the transition scene area.

3. The method for positioning autonomous vehicles in transitional scenarios using a multi-domain integrated space-ground fusion system according to claim 2, characterized in that, Step 1-2 utilize The clustering algorithm calculates the number of dominant color pixels in each sub-image block. Specifically, all pixels in the sub-image are treated as data points, and the RGB value of each pixel is used as its feature vector.

4. The method for positioning autonomous vehicles in transitional scenarios using a multi-domain integrated space-ground fusion system according to claim 2, characterized in that, Steps 1-4 employ a loop to calculate the one-dimensional dynamic programming matching result of the two principal color feature vectors in each group. and These are the principal color feature vectors of two images. Dynamic programming matching techniques are used to determine the matching distance between the image to be matched and the images in the transition scene database. When the matching distance exceeds a set threshold... At that time, it was determined that the vehicle had reached the transition scene area.

5. The method for positioning autonomous vehicles in transitional scenarios using a multi-domain integrated space-ground fusion system according to claim 1, characterized in that, The GNSS / LEO / IMU integrated positioning model is constructed based on GNSS, LEO, and IMU, using positioning information vectors output by either the GNSS or LEO positioning module. and the positioning information vector output by the IMU positioning module. Location information is fused, and represented as: The fused positioning information is then processed using a robust filtering algorithm.

6. The method for positioning a transitional scene of an autonomous vehicle using a multi-domain integrated space-ground fusion system according to claim 5, characterized in that, The switching between the GNSS positioning module and the LEO positioning module is based on the following: when the number of stable visible satellites is greater than or equal to 4, the positioning information vector output by the GNSS positioning module is used. When the number of stable visible satellites is less than 4, the positioning information vector output by LEO is used. .

7. The method for positioning autonomous vehicles in transitional scenarios using a multi-domain integrated space-ground fusion system according to claim 1, characterized in that, The LiDAR / IMU integrated positioning model is constructed based on LiDAR and IMU, and the positioning information vector is solved based on LiDAR. and the positioning information vector obtained based on IMU Location information is fused, and represented as: The fused positioning information is then processed using a robust filtering algorithm.

8. The method for positioning a transitional scene of an autonomous vehicle using a space-ground integrated multi-domain fusion system according to claim 4, characterized in that, Dynamic programming matching technique determines the matching distance between the two as follows: in, This represents the principal color feature vector of the u-th currently acquired image. This represents the main color feature vector of the image in the image database for the v-th transition scene. express and The distance between elements express The value, express The value, , , It represents the number of images.

Images