A high-precision positioning method in a transition scene based on vehicle-road cooperation

Abstract

The application provides a high-precision positioning method in a transition scene based on vehicle-road cooperation, first, a vehicle-mounted camera is used to take a picture of a road scene in front of the vehicle to obtain a to-be-matched image, a GNSS receiver is used to obtain a positioning output result of a vehicle end based on GNSS signals, and a roadside end camera is used to obtain a positioning output result of the roadside end based on vision; second, a transition scene database is constructed; then, the transition scene is identified by combining global feature matching and local feature matching based on the to-be-matched image; finally, in the transition scene area, the positioning output result of the vehicle end based on GNSS signals is smoothly switched with the positioning output result of the roadside end based on vision. The application provides a new solution for the deficiency of a traditional positioning method in a transition scene, can effectively solve the problem of insufficient positioning accuracy in the transition scene, ensures the continuity and smoothness of the trajectory of the intelligent vehicle in the driving process, and has a broad market prospect.

Description

Technical Field

[0001] This invention belongs to the field of intelligent vehicle positioning technology, specifically relating to a high-precision positioning method for transitional scenarios based on vehicle-road cooperation. Background Technology

[0002] With the rapid development of technology, intelligent vehicles have become an important component of modern transportation systems. Intelligent vehicles not only improve driving safety but also enhance traffic efficiency and reduce environmental pollution. Intelligent vehicle positioning technology, as one of its core technologies, is a crucial component of modern intelligent transportation systems and the foundation for realizing functions such as autonomous driving, route planning, and vehicle monitoring. Its development has evolved from a single technology to multi-technology integration, and then to intelligent applications. With continuous technological advancements, intelligent vehicle positioning technology will play an increasingly important role in areas such as autonomous driving, traffic management, and smart city construction.

[0003] High-precision positioning is fundamental to autonomous driving path planning. Positioning errors directly affect the accuracy and precision of planning and control algorithms. Currently, positioning technology for intelligent vehicles has made significant progress and breakthroughs. The most widely used vehicle positioning technology is the Global Navigation Satellite System (GNSS), often simply called satellite navigation. Due to its simple principle and low cost, most devices using GNSS for positioning employ single-point GNSS positioning technology, generally achieving only meter-level accuracy. In recent years, with the development of technologies such as Pecise Point Positioning (PPP) and Real-Time Kinematic (RTK), positioning accuracy has reached the centimeter level. However, due to the canyon effect, the application scenarios for positioning using GNSS technology alone are limited. In outdoor scenarios, in order to overcome the shortcomings of a single technology, a combined positioning technology based on GNSS and INS (Inertial Navigation System) has emerged. Through data fusion, positioning accuracy and reliability can be effectively improved. In addition, with the development of sensor technology, various sensors such as LiDAR, cameras, and ultrasonic sensors have been introduced into vehicle positioning systems to improve the vehicle's perception of the surrounding environment, further enhancing the accuracy and stability of positioning.

[0004] However, there is currently little research on positioning technology in indoor-outdoor transition scenarios. GNSS signals are often affected before moving from open outdoor environments to indoor environments, tunnels, underground parking garages, under overpasses, etc., which severely restricts the positioning accuracy in transition scenarios and makes it difficult to meet the high-precision positioning requirements of intelligent vehicles. Summary of the Invention

[0005] To address the shortcomings of existing technologies, this invention provides a high-precision positioning method for transitional scenarios based on vehicle-road cooperation.

[0006] The present invention achieves the above-mentioned technical objectives through the following technical means.

[0007] A high-precision positioning method for transitional scenarios based on vehicle-road cooperation includes:

[0008] First, the vehicle-mounted camera takes pictures of the road scene ahead of the vehicle to obtain images to be matched. The GNSS receiver obtains the positioning output results of the vehicle based on GNSS signals, and the roadside camera obtains the positioning output results based on roadside vision. Second, a transition scene database is constructed. Then, based on the images to be matched, the transition scenes are identified by combining global feature matching and local feature matching. Finally, in the transition scene area, the positioning output results of the vehicle based on GNSS signals and the positioning output results based on roadside vision are smoothly switched.

[0009] Furthermore, the specific steps for constructing the transition scenario database are as follows:

[0010] Along the road direction of the transition scene at the roadside end, the road scene is photographed by setting data sampling points in the transition scene, and the global feature descriptors and local feature descriptors of the sampling point images are combined to form the transition scene database.

[0011] Furthermore, the global feature descriptor is extracted using a CNN and NetVLAD framework-based method, and the local feature descriptor is obtained by extracting ORB features from the image at each sampling point.

[0012] Furthermore, the transition scenarios include transition scenarios from outdoors to indoors and transition scenarios from indoors to outdoors.

[0013] Furthermore, the global feature matching process is specifically as follows:

[0014] First, calculate the Euclidean distance between the global feature descriptor of the image to be matched and the global feature descriptors of the images in the transition scene database; then, compare the Euclidean distance between the image to be matched and all images in the transition scene database to obtain a comparison of the similarity between the images; finally, select the top N images with the highest similarity as the global feature matching result.

[0015] Furthermore, the process of local feature matching is specifically as follows:

[0016] First, local feature matching is an exact match performed on the global feature matching result, let ζ xyThe candidate image with the highest similarity to the image to be matched in the transition scene database is selected from the transition scene database. xy The eight neighboring images serve as the search space for precise matching and localization, and the search space S for all local feature matching has 9N images. Secondly, the most similar match is described by calculating the Hamming distance between the local feature descriptors of the image to be matched and the local feature descriptors of the images in the search space S. Finally, the successfully matched images are sorted in descending order of the number of correctly matched point pairs of local feature descriptors to generate a node set Q = {Q1, Q2, ..., Q...}. m}

[0017] Furthermore, the sampling point locations corresponding to node Q1 and their eight nearest neighbors in the transition scenario database are selected, and vehicle positioning information (x) based on vehicle-road cooperation is obtained through weighted fusion. l y l The calculation formula is as follows:

[0018]

[0019] In the formula, n represents the location information numbers of node Q1 and its eight nearest neighbors, n = 1, 2, ..., 9. x represents the weighting coefficient. n The x-coordinate of the image sampling point in the database is represented by y. n This represents the ordinate of the image sampling points in the database. q n K represents the nearest neighbor weighting coefficient, λ is an empirical parameter, 0 < λ < 1, and K n This represents the number of matching point pairs in the image.

[0020] Furthermore, the smooth transition includes the shift from outdoor to indoor scene positioning mode and the shift from indoor to outdoor scene positioning mode.

[0021] Furthermore, the specific process of changing the positioning method from outdoor to indoor scenes is as follows:

[0022] When transitioning from an outdoor scene to a transitional scene, the positioning output based on the vehicle's GNSS signal [x] g ,y g ] T Smooth transition to vehicle-road cooperative localization output [x] l ,y l ] T The real-time positioning coordinates (x, y) of the vehicle are obtained. t y t The calculation formula is as follows:

[0023]

[0024] in, Let λ(0) be the weight function, e be the natural constant, and let the moment when the transition scene is successfully identified be the 0th moment of switching the weight function, where λ(0) = 0;

[0025] When transitioning from a transitional scene to an indoor scene, the vehicle-road cooperative localization output result [x] is provided. l ,y l ] T Smooth transition to visual localization output based on roadside cameras [x] c ,y c ] T The real-time positioning coordinates (x, y) of the vehicle are obtained. t y t The calculation formula is as follows:

[0026]

[0027] Furthermore, the specific process of changing the positioning method from indoor to outdoor scenes is as follows:

[0028] When transitioning from an indoor scene to a transitional scene, the visual localization output based on the roadside camera [x] c ,y c ] T Smooth transition to vehicle-road cooperative localization output [x] l ,y l ] T The real-time positioning coordinates (x, y) of the vehicle are obtained. t y t The calculation formula is as follows:

[0029]

[0030] in, Let λ(0) be the weight function, e be the natural constant, and let the moment when the transition scene is successfully identified be the 0th moment of switching the weight function, where λ(0) = 0;

[0031] When transitioning from a transitional scene to an outdoor scene, the vehicle-road cooperative localization output result [x] is provided. l ,y l ] T Smooth transition to positioning output results based on vehicle-mounted GNSS signals [x] g ,y g ] T The real-time positioning coordinates (x, y) of the vehicle are obtained. t y t The calculation formula is as follows:

[0032]

[0033] The beneficial effects of this invention are as follows:

[0034] (1) In view of the problem of decreased positioning accuracy faced by traditional positioning methods in transitional scenarios, this invention proposes a vehicle-road cooperative positioning method that is different from the current mainstream vehicle-road cooperative positioning method and is suitable for transitional scenarios. It effectively solves the problem of positioning performance degradation in transitional scenarios and improves the positioning accuracy in the area.

[0035] (2) Innovatively, this invention smoothly transitions the outdoor GNSS signal-based positioning method to the indoor roadside visual positioning method in the transition scene area, realizing the positioning of intelligent vehicles in all scenarios.

[0036] (3) This invention does not rely on expensive equipment such as lidar, but uses a low-cost camera to achieve positioning and switching of transitional scenes, which can be commercially applied on a large scale. Compared with traditional positioning technology, it ensures the continuity of the trajectory of intelligent vehicles during driving and has broad market prospects. Attached Figure Description

[0037] Figure 1 This is a flowchart illustrating the high-precision positioning method for transitional scenarios based on vehicle-road cooperation as described in this invention.

[0038] Figure 2 This is a schematic diagram of the transition scenario database described in this invention;

[0039] Figure 3 This is an example of transition scene recognition as described in this invention;

[0040] Figure 4 This is a schematic diagram illustrating the vehicle positioning information obtained through weighted fusion based on vehicle-road cooperation as described in this invention. Detailed Implementation

[0041] The present invention will be further described below with reference to the accompanying drawings and specific embodiments, but the scope of protection of the present invention is not limited thereto.

[0042] The positioning experiment of this invention was completed at the entrance and exit of the underground parking garage. The technical flowchart of the specific implementation of the high-precision positioning method based on vehicle-road cooperation in transitional scenarios designed in this invention is as follows: Figure 1 As shown. The specific implementation steps are as follows:

[0043] Step 1: Use the vehicle-mounted camera to take pictures of the road scene ahead of the vehicle to obtain the image to be matched (for step 3, recognition of transition scenes); use the GNSS receiver on the vehicle to obtain latitude and longitude information and cooperate with the IMU (Inertial Measurement Unit) and RTK base station positioning data to obtain the high-precision positioning result based on the GNSS signal output of the vehicle (for step 4, switching of positioning mode); deploy a camera on the roadside to collect the beacon data information of the vehicle and solve for the high-precision positioning result based on the visual navigation information output of the roadside (for step 4, switching of positioning mode).

[0044] Step 2: Build a transition scenario database.

[0045] Step 2-1: Take photos of the road scene along the road direction of the transition scene at the roadside end by setting data sampling points in the transition scene;

[0046] Step 2-2: To further accelerate the recognition of subsequent transitional scenes, global feature extraction and local feature extraction are used to describe the similarity between scenes and store it in the roadside equipment that interacts with the vehicle.

[0047] Step 2-1 is as follows Figure 2 As shown, it is first assumed that the camera height at the sampling point is the same as the vehicle-mounted camera height, and there is no vertical relative displacement between the vehicle and the ground; the origin is taken as the position of the first image captured near the road edge, and the direction of road travel is taken as x. l axis, with x l The direction of the axis rotating 90° counterclockwise is y. l The x-axis is used to construct the coordinate system for the transition scene database. This is done by defining the x-axis within the transition scene database coordinate system. l With directional spacing a = 0.1m, in the transition scene database coordinate system y l With a directional interval b = 0.2m, sampling points are set, and a camera is used to collect image data of the transition scene;

[0048] Step 2-2 employs a global feature extraction method based on the CNN (Convolutional Neural Networks) + NetVLAD (Net Vector of Local Aggregated Descriptors) framework to obtain global feature descriptors; ORB (Oriented Fast and Rotated BRIEF) features are extracted from each sampled image to obtain local feature descriptors; further, the global and local feature descriptors of all sampled images are combined to form a transition scene database.

[0049] Step 3: Identify the transition scene: Take the image of the scene in front of the vehicle captured by the vehicle camera in Step 1 as the image to be matched, match the image to be matched with the transition scene image, search for the most similar image in the transition scene database, and calculate the real-time position parameters of the intelligent vehicle.

[0050] Step 3 identifies transitional scenes through global feature matching and local feature matching. Transitional scenes include transitional scenes from outdoors to indoors and transitional scenes from indoors to outdoors. The identification process for both is the same. Here, we will take the identification of the transitional scene from outdoors to indoors as an example.

[0051] The process of recognizing transitional scenes from outdoors to indoors:

[0052] Global feature matching performs global feature matching between the images to be matched captured in real time by the vehicle-mounted camera and the images in the transition scene database, by calculating the Euclidean distance E. uc The similarity is described by the following formula:

[0053]

[0054] In the formula, G q G represents the set of global feature descriptors for the image to be matched. j Let represent the set of global feature descriptors for images in the transition scene database, where 'i' represents the global feature descriptor number, i = 1, 2, ..., n. The smaller the Euclidean distance, the more similar the scene images are, and the closer their positions. The image to be matched is sequentially compared with all images in the transition scene database. If the Euclidean distance is less than a threshold ω, the two images are considered similar; if the Euclidean distance is greater than the threshold ω, the two images are considered dissimilar. Finally, the top N images with the highest similarity are selected as the global feature matching result; in this embodiment, N = 3.

[0055] Local feature matching further refines the matching based on the global feature matching results, let ζ xy This is the candidate image in the database that is most similar to the image to be matched, where x and y are the x and y coordinates of the image sampling points in the database, respectively. ζ is selected from the database. xy The eight surrounding images serve as the search space for further precise matching and localization, and the search space S for matching all local features has 9N images.

[0056] Based on ORB feature extraction, local features of the image to be matched and the images in the search space S are obtained. The Hamming distance is used to describe the most similar match, and the calculation method is as follows:

[0057]

[0058] In the formula, Bq B represents the set of local feature descriptors of the image to be matched. j Let S represent the set of local feature descriptors of the image in the search space S, where i represents the local feature descriptor number, i = 1, 2, ..., 256.

[0059] A threshold τ is set. If the matching result is less than the threshold, the match is considered successful; if the matching result is greater than the threshold, the match is considered unsuccessful. The Random Sample Consensus (RANSAC) algorithm is used to remove erroneous local feature descriptor pairs, and the retained pairs are the final matching pairs.

[0060] Finally, the successfully matched images are sorted in descending order of the number of matching point pairs, generating a node set Q = {Q1, Q2, ..., Q...} m This is used to determine if a vehicle has entered a transitional scene area. The transitional scene recognition results are as follows: Figure 3 As shown.

[0061] Select the sampling point location corresponding to node Q1 in the transition scenario database at this time and its eight nearest neighbors, and obtain the vehicle positioning information (x) based on vehicle-road cooperation through weighted fusion. l y l ),like Figure 4 As shown, its specific implementation method is as follows:

[0062]

[0063] In the formula, n represents the location information numbers of node Q1 and its eight nearest neighbors, n = 1, 2, ..., 9. x represents the weighting coefficient. n The x-coordinate of the image sampling point in the database is represented by y. n This represents the ordinate of the image sampling points in the database. q n K represents the nearest neighbor weighting coefficient, λ is an empirical parameter, 0 < λ < 1, and K n This represents the number of matching point pairs in the image.

[0064] Step 4: Switching the positioning method in the transition scene: When entering the transition scene, the positioning output result based on GNSS signals of the vehicle end is smoothly transitioned to the visual positioning output result based on the roadside camera; when leaving the transition scene, the visual positioning output result based on the roadside camera is smoothly transitioned to the positioning output result based on GNSS signals.

[0065] Step 4 achieves a smooth transition between indoor and outdoor scene positioning by designing a smooth switching transition positioning strategy.

[0066] (1) The shift in scene positioning methods from outdoor to indoor:

[0067] When transitioning from an outdoor scene to a transitional scene, the positioning output based on the vehicle's GNSS signal [x] g ,y g ] T Smooth transition to vehicle-road cooperative localization output [x] l ,y l ] T The real-time positioning coordinates (x, y) of the vehicle are obtained. t y t The calculation formula is as follows:

[0068]

[0069] in, Let λ(0) be the weight function, e be the natural constant, and let the moment when the transition scene is successfully identified be the 0th moment of switching the weight function, where λ(0) = 0.

[0070] When transitioning from a transitional scene to an indoor scene, the vehicle-road cooperative localization output result [x] is provided. l ,y l ] T Smooth transition to visual localization output based on roadside cameras [x] c ,y c ] T The real-time positioning coordinates (x, y) of the vehicle are obtained. t y t The calculation formula is as follows:

[0071]

[0072] This enables a shift in positioning methods from outdoor to indoor scenarios.

[0073] (2) The shift in scene positioning methods from indoor to outdoor:

[0074] When transitioning from an indoor scene to a transitional scene, the visual localization output based on the roadside camera [x] c ,y c ] T Smooth transition to vehicle-road cooperative localization output [x] l ,y l ] T The real-time positioning coordinates (x, y) of the vehicle are obtained. t y t The calculation formula is as follows:

[0075]

[0076] When transitioning from a transitional scene to an outdoor scene, the vehicle-road cooperative localization output result [x] is provided. l ,y l ] T Smooth transition to positioning output results based on vehicle-mounted GNSS signals [x] g ,y g ] T The real-time positioning coordinates (x, y) of the vehicle are obtained. t y t The calculation formula is as follows:

[0077]

[0078] This enables a shift in positioning methods from indoor to outdoor scenarios. This concludes the entire process of designing a high-precision positioning method for transitional scenarios based on vehicle-road cooperation, as described in this invention.

[0079] In summary, the high-precision positioning method for transitional scenarios based on vehicle-road cooperation designed in this invention solves the problem of decreased positioning accuracy in transitional scenarios and achieves the goal of high-precision positioning and smooth switching of intelligent vehicles in transitional scenarios.

[0080] The embodiments described above are preferred embodiments of the present invention, but the present invention is not limited to the above embodiments. Any obvious improvements, substitutions or modifications that can be made by those skilled in the art without departing from the essence of the present invention shall fall within the protection scope of the present invention.

Claims

1. A high-precision positioning method for transitional scenarios based on vehicle-road cooperation, characterized in that: First, the vehicle-mounted camera takes pictures of the road scene in front of the vehicle to obtain the image to be matched. Then, the GNSS receiver is used to obtain the positioning output result based on the GNSS signal of the vehicle. Finally, the roadside camera is used to obtain the visual positioning output result based on the roadside. Secondly, a transition scene database is constructed; then, based on the images to be matched, transition scenes are identified through a combination of global feature matching and local feature matching; specifically including: The global feature matching process is as follows: First, calculate the Euclidean distance between the global feature descriptor of the image to be matched and the global feature descriptor of the images in the transition scene database; then, compare the Euclidean distance between the image to be matched and all images in the transition scene database in turn to obtain a comparison of the similarity between the images; finally, select the top N images with the highest similarity as the global feature matching result. The local feature matching process is as follows: First, local feature matching is a precise matching performed on the global feature matching result. Let... It is the candidate image with the highest similarity to the image to be matched in the transition scene database. It is selected from the transition scene database. The eight neighboring images serve as the search space for precise matching and localization, and the search space S for all local feature matching has 9N images. Secondly, the most similar match is described by calculating the Hamming distance between the local feature descriptors of the image to be matched and the local feature descriptors of the images in the search space S. Finally, the successfully matched images are sorted in descending order of the number of correctly matched point pairs of local feature descriptors to generate a node set. ; The sampling point location corresponding to node Q1 and its eight nearest neighbors in the transition scenario database are selected, and vehicle positioning information (x) based on vehicle-road cooperation is obtained through weighted fusion. l y l The calculation formula is as follows: In the formula, n represents the location information number of node Q1 and its eight nearest neighbors, n=1,2,...,9. x represents the weighting coefficient. n The x-coordinate of the image sampling point in the database is represented by y. n This represents the ordinate of the image sampling points in the database. q n K represents the nearest neighbor weighting coefficient, λ is an empirical parameter, 0 < λ < 1, and K n This represents the number of matching point pairs in the image; Finally, in the transitional scenario area, the vehicle-side positioning output based on GNSS signals is smoothly switched to the roadside visual positioning output.

2. The high-precision positioning method for transitional scenarios based on vehicle-road cooperation according to claim 1, characterized in that, The specific steps for constructing the transition scenario database are as follows: Along the road direction of the transition scene at the roadside end, the road scene is photographed by setting data sampling points in the transition scene, and the global feature descriptors and local feature descriptors of the sampling point images are combined to form the transition scene database.

3. The high-precision positioning method for transitional scenarios based on vehicle-road cooperation according to claim 2, characterized in that, The global feature descriptor is extracted using a CNN and NetVLAD framework-based method, while the local feature descriptor is obtained by extracting ORB features from the image at each sampling point.

4. The high-precision positioning method for transitional scenarios based on vehicle-road cooperation according to claim 1, characterized in that, The transition scenarios include transition scenarios from outdoors to indoors and transition scenarios from indoors to outdoors.

5. A high-precision positioning method for transitional scenarios based on vehicle-road cooperation according to claim 1, characterized in that, The smooth transition includes the shift from outdoor to indoor scene positioning methods and the shift from indoor to outdoor scene positioning methods.

6. A high-precision positioning method for transitional scenarios based on vehicle-road cooperation according to claim 5, characterized in that, The specific process of changing the positioning method from outdoor to indoor scenes is as follows: When transitioning from an outdoor scene to a transitional scene, the positioning output based on the vehicle's GNSS signal... Smooth transition to vehicle-road cooperative localization output results The real-time positioning coordinates (x, y) of the vehicle are obtained. t y t The calculation formula is as follows: in, Let be the weighting function, e be the natural constant, and t be the time variable with the successful recognition of the transition scene as the zero point. Let the successful recognition of the transition scene be the zero point of the switching weighting function. ; When transitioning from a transitional scene to an indoor scene, the vehicle-road cooperative positioning output results are provided. Smooth transition to visual localization output based on roadside cameras The real-time positioning coordinates (x, y) of the vehicle are obtained. t y t The calculation formula is as follows: 。 7. A high-precision positioning method for transitional scenarios based on vehicle-road cooperation according to claim 5, characterized in that, The specific process of changing the positioning method from indoor to outdoor scenes is as follows: When transitioning from an indoor scene to a transitional scene, the visual positioning output based on the roadside camera... Smooth transition to vehicle-road cooperative localization output results The real-time positioning coordinates (x, y) of the vehicle are obtained. t y t The calculation formula is as follows: in, Let be the weighting function, e be the natural constant, and t be the time variable with the successful recognition of the transition scene as the zero point. Let the successful recognition of the transition scene be the zero point of the switching weighting function. ; When transitioning from a transitional scene to an outdoor scene, the vehicle-road cooperative positioning output results are provided. Smooth transition to positioning output results based on vehicle-mounted GNSS signals The real-time positioning coordinates (x, y) of the vehicle are obtained. t y t The calculation formula is as follows: 。

Images