Intelligent vehicle lateral positioning method based on low earth orbit satellite enhanced GNSS and visual multi-domain fusion
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- JIANGSU UNIV
- Filing Date
- 2025-03-24
- Publication Date
- 2026-05-12
AI Technical Summary
Existing technologies struggle to achieve high-precision lateral positioning for intelligent vehicles in complex traffic scenarios, especially in densely built-up areas and signal blind spots such as tunnels, where the positioning accuracy of satellite navigation systems is limited and a single data source cannot meet the requirements.
By combining positioning and navigation information from low-orbit satellite-enhanced GNSS, a reference high-precision map is created. Through map matching using a weighted model, lane lines are detected visually, and the lateral position is calculated using a Gaussian Bayes model, achieving high-precision lateral positioning for intelligent vehicles.
It improves global positioning accuracy, simplifies the map matching process, improves path confirmation accuracy, reduces the impact of sensor motion state changes on lateral positioning, and achieves high-precision intelligent vehicle lateral positioning.
Smart Images
Figure CN120213068B_ABST
Abstract
Description
Technical Field
[0001] This invention relates to the fields of satellite positioning and computer vision, specifically to a method for lateral positioning of intelligent vehicles based on low-orbit satellite-enhanced GNSS and visual multi-domain fusion. Background Technology
[0002] Advances in science and technology are continuously driving the intelligent development of automobiles, and high-precision positioning is a crucial component of this development. For intelligent vehicles, highly accurate location information is a prerequisite for perceiving traffic conditions, planning driving routes, and making driving decisions. As the foundation for the normal operation of autonomous driving systems, the demand for high-precision positioning in intelligent vehicles is increasing with the continuous development and popularization of autonomous driving technology, especially in lateral positioning. Lateral positioning allows intelligent vehicles to not only determine their current global location but also obtain the specific lane they are on and their precise position within that lane, enabling more accurate route planning and navigation. Currently, lateral positioning for intelligent vehicles mainly relies on the integrated application of technologies such as satellite navigation systems (e.g., GPS, BeiDou), onboard sensors, and high-precision maps.
[0003] Satellite navigation systems are currently the most convenient and efficient means of positioning for intelligent vehicles. By receiving satellite signals, vehicles can determine their latitude, longitude, and altitude, thus achieving positioning on a map. However, satellite signals are easily affected by the canyon effect in areas with dense high-rise buildings, and signal blind spots exist in areas such as tunnels, limiting the accuracy and range of positioning. Low Earth Orbit (LEO) satellites, orbiting at altitudes below 2000 km above the Earth's surface, have advantages over medium and high Earth Orbit satellites, including stronger signal strength and richer spectrum resources, effectively enhancing receiver signal quality and anti-interference capabilities. Therefore, LEO satellites can effectively compensate for the shortcomings of GNSS positioning, making high-precision lateral positioning possible in more scenarios. Simultaneously, the application of onboard sensors can collect environmental information around the vehicle, such as road information, providing more reference data for lateral positioning. Furthermore, high-precision maps are also an indispensable part of high-precision positioning for intelligent vehicles. High-precision maps can provide detailed road information, including lane lines, traffic signs, and intersection structures, thereby improving positioning accuracy and driving safety.
[0004] Currently, a single data source is often insufficient to meet the high-precision lateral positioning requirements in complex traffic scenarios. Summary of the Invention
[0005] To address the shortcomings of existing technologies, this invention proposes a lateral positioning method for intelligent vehicles based on low-orbit satellite-enhanced GNSS and visual multi-domain fusion. This invention utilizes positioning and navigation information from low-orbit satellite-enhanced GNSS and provides a low-cost reference map creation method. The driving path is obtained by performing map matching based on a weighted model on the reference map, and the specific lane of the current path is determined by visual detection of lane lines. Finally, based on a Gaussian Bayes model, the lateral position within the current lane is calculated, achieving high-precision lateral positioning of the intelligent vehicle.
[0006] The technical solution adopted in this invention is as follows:
[0007] The intelligent vehicle lateral positioning method based on low-orbit satellite-enhanced GNSS and visual multi-domain fusion includes the following steps:
[0008] Step 1: Based on the precise point positioning model of LOR satellite-enhanced GNSS, perform pseudorange point positioning and precise point positioning data processing on LOR-enhanced GNSS to obtain the latitude and longitude coordinates after enhanced positioning;
[0009] Step 2: Create a reference high-precision map, perform map matching based on a weighted model on the reference high-precision map, and filter to obtain the initial path;
[0010] Step 3: Based on the initial path obtained through filtering, use visual detection of lane lines to determine the specific lane the intelligent vehicle is currently in by the positional relationship of the lane lines within the image.
[0011] Step 4: Based on the confirmed specific lane, use a Gaussian Bayes model to calculate the lateral position of the intelligent vehicle in the current lane.
[0012] Furthermore, the precise point positioning model using low-Earth orbit satellites to enhance GNSS is denoted as:
[0013]
[0014] Where G represents GNSS and L represents LEO satellite, This indicates the location of ionospherically non-spherically enhanced pseudorange observations using low-Earth orbit satellites. This represents the combined ionospheric phase observations enhanced by low-Earth orbit satellites. This indicates the satellite-to-ground distance enhanced by low-orbit satellites. The λ represents the hardware delay correction for ionospheric pseudorange, where T represents the tropospheric delay. L Indicates the wavelength of LEO satellites without ionosphere. Indicates the ambiguity of the non-ionospheric combination. This represents the sum of all other errors in the pseudorange. This is the sum of all other phase errors. This indicates the deviation between different systems.
[0015] Furthermore, the method for creating a reference high-precision map is as follows: based on high-resolution satellite remote sensing images or maps from map service providers, use map editing software to build a reference map road network.
[0016] Furthermore, the map matching method is as follows:
[0017] S1. After obtaining the latitude and longitude information of the intelligent vehicle, calculate the Euclidean distance d from the location to the surrounding paths;
[0018] S2. Determine the driving direction and path direction of the intelligent vehicle, and further filter to obtain the initial path based on the matching degree of the driving direction and path direction;
[0019] S3. Integrate the screening methods of S1 and S2 to construct a weighted model:
[0020] Result = 0.4d * +0.6m *
[0021] In the formula, d * m represents the distance parameter after d is normalized. * This represents the normalized direction matching parameter.
[0022] Furthermore, step 3 utilizes a visual method to detect lane lines as follows:
[0023] Lane images are acquired using an onboard camera and then input into the Ultra-Fast-Lane-Detection-V2 deep learning algorithm for lane line detection.
[0024] Furthermore, the method for estimating the lateral position of the vehicle within the current lane using a Gaussian-Bayes model is as follows:
[0025] Step 4.1: Extract the obtained lane lines, obtain the two edges of each lane line in the current lane, and construct the geometric constraints in the ground plane;
[0026] Step 4.2: The Gaussian-Bayes model transforms the motion state estimation problem of the vehicle camera into the problem of determining the optimal ground plane normal vector, and derives the formula for determining the plane normal vector to determine the position of the ground plane normal vector.
[0027] Step 4.3: After determining the position of the normal vector of the ground plane, establish a ground plane reference coordinate system based on the vector direction; after the ground plane coordinate system is determined, use the rotation matrix R and translation vector t to establish the transformation relationship between the image pixel coordinates and the ground plane coordinates; after the coordinate transformation, calculate the distance from the origin of the ground plane coordinate system to the lane line to calculate the lateral position in the current lane, and use the fixed width of the lane line as a scaling factor to restore the distance to the true size.
[0028] Furthermore, in step 4.1, Canny edge detection and weighted least squares fitting are used to extract the two edge lines of each lane line in detail.
[0029] Furthermore, geometric constraints are constructed in the ground plane: the distance between the two edge lines of the same lane line is fixed, and there is a parallel relationship between the edge lines, thus constructing geometric constraints in the ground plane.
[0030] Furthermore, the transformation relationship between image pixel coordinates and ground plane coordinates is established using the rotation matrix R and the translation vector t:
[0031]
[0032] In the formula, u and v are the x-coordinate and y-coordinate of the image pixel, respectively, s is the scale factor, K is the camera intrinsic parameter matrix, r1 and r2 are the first and second column elements of the rotation matrix R, t is the translation vector, and X and Y are the x-coordinate and y-coordinate in the ground plane coordinate system, respectively.
[0033] Furthermore, the formula for determining the plane normal vector is denoted as:
[0034]
[0035] Where n is the number of ground geometric constraints, d i (N) represents the error associated with the plane normal vector N, q i These are known, definite values associated with this geometric property.
[0036] The beneficial effects of this invention are:
[0037] With the gradual deployment of low-Earth orbit satellite constellations, integrated space-ground multi-domain positioning offers the possibility of solving lateral positioning problems in complex traffic scenarios. Therefore, this paper aims to provide a high-precision lateral positioning scheme for locations and areas with poor GNSS signal positioning. The technical effects of the specific solution are as follows:
[0038] 1. Use low-orbit satellites to enhance GNSS signals and improve global positioning accuracy.
[0039] 2. Using reference points to represent paths and intersections in the reference map, a map matching method based on a weighted model is proposed, which simplifies the map matching process and improves the accuracy of path confirmation.
[0040] 3. Within the path, the specific lane is identified by the geometric positional relationship of the lane lines in the image, which is simple and efficient.
[0041] 4. The lateral position is calculated based on the Gaussian Bayes model within the main lane, which reduces the impact of sensor motion state changes on lateral distance calculation and improves the accuracy of lateral positioning.
[0042] 5. Research on using low-orbit satellites to enhance GNSS signals to improve global positioning accuracy. Based on this positioning information, road matching is performed on a reference high-precision map and fused with visual information to achieve accurate lane confirmation. Finally, based on the Gaussian Bayes model, the lateral position of the intelligent vehicle in the current lane is calculated to achieve high-precision lateral positioning of the intelligent vehicle through multi-domain fusion of space, air, and ground. Attached Figure Description
[0043] Figure 1 This is the overall flowchart of this method.
[0044] Figure 2 A schematic diagram of GNSS enhancement for low-Earth orbit satellites.
[0045] Figure 3 Create a road matching diagram for the reference map.
[0046] Figure 4 This is a schematic diagram confirming the specific lanes within the route.
[0047] Figure 5 This is a schematic diagram for calculating the lateral position within the lane. Detailed Implementation
[0048] 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.
[0049] To achieve the above objectives, the present invention provides a method for lateral positioning of intelligent vehicles based on low-orbit satellite-enhanced GNSS and visual fusion, comprising the following steps:
[0050] Step 1: Low-Earth Orbit (LEO) satellite augmented GNSS positioning. Based on the precise point positioning model of LEO satellite augmented GNSS, pseudorange point positioning and precise point positioning data processing are performed on LEO satellite augmented GNSS. The observation data, precise orbit data, and antenna file data are all obtained through simulation. Finally, the latitude and longitude coordinates after augmented positioning are obtained.
[0051] The principle of LEO satellite-enhanced GNSS positioning is as follows: In recent years, LEO satellites have rapidly developed due to their commercial advantages such as light weight, low cost, and short development cycle. The sheer number of LEO satellites means that a single receiving station can simultaneously observe far more satellites than GNSS satellites. Furthermore, their orbital altitudes, directions of motion, and signal frequencies differ, allowing for more information to be obtained for iterative calculations during positioning. This complements GNSS, and the satellite constellation offers unique advantages. Firstly, the low orbital altitude of LEO satellites results in lower launch costs and stronger transmitted impedance signals, leading to better anti-interference and anti-spoofing performance, which improves positioning performance in obscured environments. Secondly, the high speed of LEO satellites allows them to pass over the station within minutes, resulting in rapid geometric changes and shortening the convergence time for precise point positioning. Currently, LEO satellite and GNSS joint positioning technology has received widespread attention. Utilizing the advantages of LEO satellites to enhance the service performance of GNSS navigation systems is a current development trend. A diagram illustrating the orbital relationship and fusion between LEO and conventional satellites is attached. Figure 2 As shown.
[0052] The pseudorange observation model for low-Earth orbit satellites is as follows:
[0053]
[0054] In the formula, r represents the receiver and s represents the satellite. Indicates the pseudo-range of the non-ionospheric combination. dt represents the phase observation value. r dt represents the clock difference at the corresponding end of the receiver. s Indicates satellite clock bias, b r b represents the pseudorange hardware delay of the ionosphere-free combined receiver. s Indicates the pseudorange hardware delay of ionospheric combined satellites; δ r δ represents the phase hardware delay of the ionosphere-free combined receiver. s This indicates the phase hardware delay of the ionospheric composite satellite. Indicates the satellite-to-Earth distance, λ represents the non-ionospheric combination wavelength, and N... s Indicates the ambiguity of the non-ionospheric combination. This represents the sum of all other errors in the pseudorange. This represents the sum of all other phase errors, where c is the speed of light.
[0055] The precise point positioning model using low-Earth orbit satellites to enhance GNSS is as follows:
[0056]
[0057] Where G represents GNSS and L represents LEO satellite, This indicates the location of ionospherically non-spherically enhanced pseudorange observations using low-Earth orbit satellites. This represents the combined ionospheric phase observations enhanced by low-Earth orbit satellites. This indicates the satellite-to-ground distance enhanced by low-orbit satellites. λ represents the hardware delay correction for GNSS ionospheric pseudorange, where T represents the tropospheric delay. L Indicates the wavelength of LEO satellites without ionosphere. Indicates the ambiguity of the non-ionospheric combination. This represents the sum of all other errors in the pseudorange. This is the sum of all other phase errors. The formula for representing the deviation between different systems is as follows:
[0058]
[0059] In the formula, TO represents the hardware delay correction for ionospheric pseudorange of LEO satellites. L,G This indicates the inherent time difference in clock reference constraints between GNSS and LEO satellites. This indicates the pseudorange hardware delay of LEO satellites without ionosphere. This indicates the hardware delay of the ionospheric-free GNSS pseudorange.
[0060] Step 2: Create a reference high-precision map, and perform map matching based on a weighted model on the reference high-precision map to obtain the initial path. Details are as follows:
[0061] In the lateral positioning of intelligent vehicles, matching based on reference map road networks is a crucial step. This invention proposes a simple method for creating and matching reference maps. Based on high-resolution satellite remote sensing images or maps from map service providers, professional map editing software is used to construct the reference map road network. Reference points are created at the beginning and end of straight paths; the straight line formed by two points represents the path. At curved lanes, the curved lane is divided into multiple short straight lines. At intersections, the reference point trajectory is refined, with a denser network of reference points to address common matching errors at intersections. The reference point format constituting the trajectory is as follows:
[0062] r = (r x ,r y (4)
[0063] Where, r x r y This indicates the longitude and latitude coordinates of the reference point.
[0064] As attached Figure 3As shown, after creating the reference map, the more accurate latitude and longitude information obtained in step 1 above is used to perform map matching for the intelligent vehicle. The specific process is as follows:
[0065] S1. After obtaining the latitude and longitude information of the intelligent vehicle, first determine the path within a certain range around the location, calculate the Euclidean distance d from the point to the straight line. Since the latitude and longitude information of the reference points at both ends of the path is known, the equation of the straight line of the path can be calculated, and then directly calculated according to the following formula:
[0066]
[0067] In the formula, A, B, and C represent the parameters of the straight line equation, and x0 and y0 represent the current latitude and longitude coordinates of the intelligent vehicle.
[0068] S2. For matching errors caused by the proximity of paths, since the intelligent vehicle's historical location is recorded very frequently, the driving direction of the intelligent vehicle can be determined based on the first few historical location records. Simultaneously, the path direction can be calculated based on the initial and final reference points. Therefore, the directional matching degree can be determined based on the driving direction and path direction to further filter and obtain the initial path.
[0069] The method for determining the direction matching degree is as follows: calculate the current angle between the current driving direction and the path direction, and normalize the current angle to obtain m. * The current included angle is divided by 180° to obtain m. * .
[0070] S3. Integrate the two screening methods into a weighted model, and in actual matching, place more trust on the matching degree of the direction:
[0071] Result = 0.4d * +0.6m * (6)
[0072] In the formula, d * m represents the normalized distance parameter. * This represents the normalized direction matching parameter.
[0073] Step 3: Based on the initial path obtained through filtering, use visual detection of lane lines to determine the specific lane the intelligent vehicle is currently in by the positional relationship of the lane lines within the image.
[0074] Based on the map matching in step 2 above, the current driving path of the intelligent vehicle can be obtained. However, different lanes on the same driving path cannot be identified using the above method because they are too close together and face the same direction. Therefore, the vehicle's camera is used to detect lane lines visually, and the specific lane the intelligent vehicle is currently in is determined by the geometric positional relationship of the lane lines in the image, as shown in the attached figure. Figure 4 As shown. Traditional visual lane detection cannot cope with changes in environment and lighting, and it also cannot accurately detect lane lines in cases of partial occlusion. Therefore, the Ultra-Fast-Lane-Detection-V2 deep learning algorithm is used for lane detection. This model uses a hybrid anchor approach to locate and detect lane lines, which can effectively cope with changes in environment and lighting, as well as lane line occlusion. Roads have multiple lanes, usually represented by two lines on the left and right sides. Road markings that separate lanes are also represented by multiple lines. The lane separated by the two lines closest to the bottom center of the image is the lane the intelligent vehicle is currently in. For high-precision navigation, it is also necessary to determine which lane the vehicle is in, which can be determined by the number of lane lines in the image.
[0075] n * =n-1 (7)
[0076] Where, n * Let n be the number of lanes and n be the number of lane lines.
[0077] Step 4: Based on the confirmed lane, calculate the lateral position of the intelligent vehicle within the current lane using a Gaussian Bayes model; the specific process is as follows:
[0078] From matching the correct path to confirming the specific lane, the final step is to calculate the lateral position within the current lane to complete the entire lateral positioning process. In existing solutions, the lateral position is calculated using a unit conversion factor between actual distance and pixels, or by constructing a computational model based on the mounting pose of the vehicle-mounted camera. These methods do not consider the complex and changing road conditions when calculating the vehicle's position. During vehicle movement, vehicle vibration, road bumps, and changes in slope all affect the camera's motion parameters, thus impacting positioning accuracy. Therefore, this invention uses a Gaussian-Bayesian model to accurately estimate the vehicle's lateral position within the current lane. The specific process is as follows:
[0079] Step 4.1 First, the obtained lane lines are extracted in a more refined manner to obtain the two edges of each lane line in the current lane and construct the geometric constraints in the ground plane.
[0080] Step 4.2: The Gaussian-Bayes model transforms the motion state estimation problem of the vehicle-mounted camera into the problem of determining the optimal ground plane normal vector. In this model, the problem of uniquely determining the normal vector of a certain plane is considered as maximizing the probability of satisfying a set of geometric constraints.
[0081]
[0082] In the formula, Let N be the optimal ground plane normal vector, and N be a given plane normal vector. Let G be the sampling plane normal vector, C be the plane geometric constraint, n be the number of plane geometric constraints, and G be the number of plane geometric constraints. i (N) represents the geometric property calculated from the given plane normal vector N, q i For a known, definite value associated with this geometric property, d i (N) represents the error associated with the plane normal vector N.
[0083] By transforming equation (8) using Bayes' theorem, the final maximum likelihood solution model is:
[0084]
[0085] In the formula, P(C i |N) represents the probability that the given sampling plane normal vector N satisfies the geometric constraints of the i-th plane.
[0086] Using the Gaussian distribution function model to analyze P(C) i Modeling is performed using |N), where the standard deviation of the Gaussian model for all constraints is equal. In practical applications, an appropriate value can be chosen, ultimately simplifying the likelihood model for each constraint:
[0087]
[0088] In the formula, G is the normalized Gaussian distribution function, σ is the standard deviation of the Gaussian model, and exp is the exponential function.
[0089] Combining the above formulas, we can finally derive the formula for determining the plane normal vector:
[0090]
[0091] The position of the normal vector of the ground plane can be determined based on the above equation (11).
[0092] Step 4.3: After determining the position of the normal vector of the ground plane, establish a ground plane reference coordinate system based on the vector direction. After the ground plane coordinate system is determined, homography exists between planes. Therefore, the transformation relationship between image pixel coordinates and ground plane coordinates is established using the rotation matrix R and the translation vector t.
[0093]
[0094] In the formula, u and v are the x-coordinate and y-coordinate of the image pixel, respectively, s is the scale factor, K is the camera intrinsic parameter matrix, r1 and r2 are the first and second column elements of the rotation matrix R, t is the translation vector, and X and Y are the x-coordinate and y-coordinate in the ground plane coordinate system, respectively.
[0095] As attached Figure 5As shown, after coordinate transformation, the lateral position within the current lane can be calculated by measuring the distance from the origin of the ground plane coordinate system to the lane line. Then, the distance can be restored to its true size by using the fixed width of the lane line as a scaling factor.
[0096] More specifically, the method for establishing a ground plane reference coordinate system based on vector direction is as follows: taking the endpoint of the ground plane normal vector as the origin of the ground plane reference coordinate system, the translation vector t between the ground plane coordinate system and the camera coordinate system can be represented by the three-dimensional coordinates of the ground plane normal vector. Since the orientation of the ground plane normal vector can be represented by the angle with the coordinate axes, the z-axis of the camera coordinate system can be rotated to the direction of the ground plane normal vector by multiple rotations around the axis (rotation matrix R). Then, by translating according to the translation matrix t, the position and direction of the X-axis and Y-axis on the ground plane can be determined, thus establishing the ground plane reference coordinate system.
[0097] The following section provides a further explanation of the specific operation process in a particular scenario:
[0098] First, follow step S1 as described in the instruction manual, as shown in the attached document. Figure 2 As shown, pseudorange point positioning and precise point positioning data processing are performed on low-Earth orbit (LEO) satellite-enhanced GNSS based on the precise point positioning model. The observation data, precise orbit data, and antenna file data are all obtained through simulation, and finally, the latitude and longitude coordinates after enhanced positioning are obtained.
[0099] As described in step S2 of the instruction manual, a reference map road network is created using professional map editing software based on high-resolution satellite remote sensing images or maps from map service providers. (See attached...) Figure 3 As shown, reference points are created at the beginning and end of a regular straight path, and the straight line formed by the two points is used to represent the path. At curved lanes, the curved lane is divided into multiple short straight segments. At intersections, the reference point trajectories are refined to create a denser road network, addressing common matching errors at intersections. After the reference map is created, map matching is performed on the reference map based on the acquired enhanced positioning information. Candidate paths are searched within a 10-meter radius centered on the current positioning point. The distance from the current positioning point to all candidate paths is calculated. Simultaneously, the current driving direction is calculated based on historical positioning points, and the angle between the current driving direction and the path direction is calculated. After obtaining the path distance and the angle between the path direction and the path direction, they are normalized, and the matching degree of the candidate path is calculated according to a weighted model. After calculating the matching degree for all candidate paths, the path with the highest matching degree is selected as the correct path, and the current positioning point is vertically projected onto this path. This projected point is the corrected positioning point on the correct path.
[0100] After correcting the location points to the correct path, lane line detection is performed using the Ultra-Fast-Lane-Detection-V2 deep learning algorithm, as described in step S3 of the instruction manual, as shown in the attached figure. Figure 4 As shown, the number of lane lines in the image is identified, the detected lane lines are numbered in order from left to right, and the lanes in the image are divided into lane 1, lane 2, and lane 3 from left to right. Since the lane separated by the two lines closest to the bottom center of the image is the lane where the intelligent vehicle is currently located, lane 2 is identified as the specific lane in the current path of the intelligent vehicle.
[0101] After confirming the current lane, following step S4 in the manual, the two lane lines obtained in step S3 are first extracted more precisely. Using Canny edge detection and weighted least squares fitting, the two edge lines of each lane line are extracted. The distance between the two edge lines of the same lane line is fixed, and the edge lines are parallel, thus constructing geometric constraints in the ground plane. Next, based on the Gaussian-Bayes model, the ground plane normal vector is uniformly sampled, and the likelihood value of the geometric constraints on each sampled plane is calculated. Maximum likelihood estimation is then performed to obtain the optimal ground plane normal vector. The translation vector t is represented by the three-dimensional coordinates of the ground plane normal vector. The z-axis of the camera coordinate system is rotated to the direction of the ground plane normal vector through multiple rotations around the axis, obtaining the rotation matrix R. Then, translation is performed based on the translation matrix t to determine the reference coordinate system on the ground plane. Based on the rotation matrix R and translation vector t, a homography transformation relationship between the ground plane and the camera plane is constructed. The pixel coordinates of the lane lines are transformed to the ground plane, and the distance from the origin of the ground plane coordinate system to the lane lines is calculated. Since the width of the lane lines is fixed, the lane line width is used as a scaling factor to restore the true distance value, thereby determining the lateral position within the current lane.
[0102] 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 lateral positioning of intelligent vehicles based on low-orbit satellite-enhanced GNSS and visual multi-domain fusion, characterized in that, Includes the following steps: Step 1: Based on the precise point positioning model of LOR satellite-enhanced GNSS, perform pseudorange point positioning and precise point positioning data processing on LOR-enhanced GNSS to obtain the latitude and longitude coordinates after enhanced positioning; Step 2: Create a reference high-precision map, and perform map matching based on a weighted model on the reference high-precision map to obtain the initial path; the map matching method is as follows: S1. After obtaining the latitude and longitude information of the intelligent vehicle, calculate the Euclidean distance d from the location to the surrounding paths; S2. Determine the driving direction and path direction of the intelligent vehicle, and further filter to obtain the initial path based on the matching degree of the driving direction and path direction; S3. Integrate the screening methods of S1 and S2 to construct a weighted model: ; In the formula, This represents the distance parameter after d is normalized. This represents the normalized orientation matching parameter; Step 3: Based on the initial path obtained through filtering, use visual detection of lane lines to determine the specific lane the intelligent vehicle is currently in by the positional relationship of the lane lines within the image. Step 4: Based on the confirmed lane, calculate the lateral position of the intelligent vehicle within the current lane using a Gaussian Bayes model; the method for estimating the lateral position of the vehicle within the current lane using a Gaussian Bayes model is as follows: Step 4.1: Extract the obtained lane lines, obtain the two edges of each lane line in the current lane, and construct the geometric constraints in the ground plane; Step 4.2: The Gaussian Bayes model transforms the motion state estimation problem of the vehicle camera into the problem of determining the optimal ground plane normal vector, and derives the formula for determining the plane normal vector to determine the position of the ground plane normal vector. Step 4.3: After determining the position of the normal vector of the ground plane, establish a ground plane reference coordinate system based on the vector direction; after the ground plane coordinate system is determined, use the rotation matrix R and the translation vector t to establish the transformation relationship between the image pixel coordinates and the ground plane coordinates; after the coordinate transformation, calculate the distance from the origin of the ground plane coordinate system to the lane line, that is, realize the calculation of the lateral position in the current lane, and use the fixed width of the lane line as a scaling factor to restore the distance to the true size.
2. The intelligent vehicle lateral positioning method based on low-orbit satellite-enhanced GNSS and visual multi-domain fusion according to claim 1, characterized in that, The precise point positioning model using low-Earth orbit satellites to enhance GNSS is denoted as: ； ； （2） Where G represents GNSS and L represents LEO satellite, This indicates the location of ionospherically non-spherically enhanced pseudorange observations using low-Earth orbit satellites. This represents the combined ionospheric phase observations enhanced by low-Earth orbit satellites. This indicates the satellite-to-ground distance enhanced by low-orbit satellites. This represents the hardware delay correction for ionospheric pseudorange. Indicates tropospheric delay, Indicates the wavelength of LEO satellites without ionosphere. Indicates the ambiguity of the non-ionospheric combination. This represents the sum of all other errors in the pseudorange. This is the sum of all other phase errors. Indicates the deviation between different systems. It is the speed of light.
3. The intelligent vehicle lateral positioning method based on low-orbit satellite-enhanced GNSS and visual multi-domain fusion according to claim 1, characterized in that, The method for creating a reference high-precision map is as follows: based on high-resolution satellite remote sensing images or maps from map service providers, use map editing software to build the reference map road network.
4. The intelligent vehicle lateral positioning method based on low-orbit satellite-enhanced GNSS and visual multi-domain fusion as described in claim 1, characterized in that, Step 3 uses a visual method to detect lane lines: acquire lane images using an onboard camera, and input the lane images into the Ultra-Fast-Lane-Detection-V2 deep learning algorithm for lane line detection.
5. The intelligent vehicle lateral positioning method based on low-orbit satellite-enhanced GNSS and visual multi-domain fusion according to claim 1, characterized in that, In step 4.1, Canny edge detection and weighted least squares fitting are used to extract the two edge lines of each lane line in detail.
6. The intelligent vehicle lateral positioning method based on low-orbit satellite-enhanced GNSS and visual multi-domain fusion according to claim 1, characterized in that, Constructing geometric constraints in the ground plane: The distance between the two edge lines of the same lane line is fixed, and there is a parallel relationship between the edge lines, thus constructing geometric constraints in the ground plane.
7. The intelligent vehicle lateral positioning method based on low-orbit satellite-enhanced GNSS and visual multi-domain fusion according to claim 1, characterized in that, The transformation relationship between image pixel coordinates and ground plane coordinates is established using the rotation matrix R and the translation vector t: ； In the formula, , These are the x and y coordinates of the image pixels, respectively. As a scale factor, For the camera intrinsic parameter matrix, These are the first and second column elements of the rotation matrix R, respectively. It is a translation vector. , These are the x-coordinate and y-coordinate in the ground plane coordinate system, respectively.
8. The intelligent vehicle lateral positioning method based on low-orbit satellite-enhanced GNSS and visual multi-domain fusion according to claim 1, characterized in that, The formula for determining the normal vector of a plane is denoted as: ； in, The number of ground geometric constraints. The geometric properties calculated from a given plane normal vector N represent the error associated with the plane normal vector N. , For known geometric properties The relevant definite values, This is the normal vector of the sampling plane.