CN-120213068-B - Intelligent automobile transverse positioning method based on low-orbit satellite enhanced GNSS and visual multi-domain fusion

CN120213068BCN 120213068 BCN120213068 BCN 120213068BCN-120213068-B

Abstract

The invention discloses an intelligent automobile transverse positioning method based on low-orbit satellite enhanced GNSS and visual multi-domain fusion, which is characterized by comprising the steps of performing pseudo-range single-point positioning and precise single-point positioning data processing on a low-orbit enhanced GNSS according to a precise single-point positioning model of the low-orbit satellite enhanced GNSS to obtain longitude and latitude coordinates after enhanced positioning, creating a reference high-precision map, performing map matching based on a weight model on the reference high-precision map, screening to obtain an initial path, obtaining the initial path based on the screening, utilizing a visual detection lane line, confirming a specific lane where an intelligent automobile is currently located through the position relation of the lane line in an image, and calculating the transverse position of the intelligent automobile in the current lane based on the confirmed specific lane by utilizing a Gao Sibei leaf model. The invention is based on low-orbit satellite enhanced GNSS and visual multi-domain fusion, and aims to provide a high-precision transverse positioning scheme in places and areas with poor GNSS signal positioning.

Inventors

LI DAICHENG
ZHONG WEI
CAI YINGFENG
CHEN LONG
WANG HAI
LIU ZE
JIANG JIN

Assignees

江苏大学

Dates

Publication Date: 20260512
Application Date: 20250324

Claims (8)

1. The intelligent automobile transverse positioning method based on the low-orbit satellite enhanced GNSS and visual multi-domain fusion is characterized by comprising the following steps of: step 1, pseudo-range single-point positioning and precise single-point positioning data processing are carried out on a low-orbit enhanced GNSS according to a precise single-point positioning model of the low-orbit satellite enhanced GNSS, and longitude and latitude coordinates after enhanced positioning are obtained; step 2, creating a reference high-precision map, and carrying out map matching based on a weight model on the reference high-precision map to screen and obtain an initial path, wherein the map matching method comprises the following steps: s1, after longitude and latitude information of an intelligent automobile is obtained, calculating Euclidean distance d from the position to surrounding paths; S2, determining the running direction and the path direction of the intelligent automobile, and further screening to obtain an initial path according to the matching degree of the running direction and the path direction; S3, integrating the screening modes of the S1 and the S2, and constructing a weight model: ; In the formula, Represents the distance parameter after d is normalized, Representing the normalized direction matching degree parameter; Step 3, obtaining an initial path based on screening, and utilizing vision to detect lane lines, and confirming a specific lane where the intelligent automobile is currently located through the position relation of the lane lines in the image; And 4, calculating the transverse position of the intelligent automobile in the current lane by utilizing a Gao Sibei phyllus model based on the confirmed specific lane, wherein the method for estimating the transverse position of the automobile in the current lane by utilizing a Gao Sibei phyllus model comprises the following steps: Step 4.1, extracting the obtained lane lines, obtaining two edges of each lane line in the current lane, and constructing geometric constraints in a ground plane; step 4.2, gao Sibei the leaf model converts the motion state estimation problem of the vehicle-mounted camera into a problem of confirming the optimal normal vector of the ground plane, deduces a formula for confirming the normal vector of the plane and is used for confirming the normal vector position of the ground plane; And 4.3, after the normal vector position of the ground plane is determined, a ground plane reference coordinate system is established according to the vector direction, after the ground plane coordinate system is determined, a conversion relation between image pixel coordinates and the ground plane coordinates is established by utilizing a rotation matrix R and a translation vector t, after coordinate conversion, the distance from the origin of the ground plane coordinate system to the lane line is calculated, namely the calculation of the transverse position in the current lane is realized, the fixed width of the lane line is used as a scale factor, and the distance is restored to the real size.
2. The intelligent automobile transverse positioning method based on low-orbit satellite-enhanced GNSS and visual multi-domain fusion according to claim 1, wherein the precise single-point positioning model using the low-orbit satellite-enhanced GNSS is recorded as follows: ; ; (2) Where G denotes the GNSS, L denotes the LEO satellite, Representing ionosphere-free combined pseudorange observations enhanced with low-orbit satellites, Representing low-orbit satellite-enhanced ionosphere-free combined phase observations, Indicating enhanced satellite distances with low-orbit satellites, Representing ionosphere-free pseudorange hardware delay corrections, Indicating the tropospheric delay is to be taken, Indicating ionosphere free combined LEO satellite wavelengths, Indicating that there is no ionospheric combined ambiguity, Representing the sum of the other errors of the pseudoranges, Is the sum of the other errors of the phase, Representing the deviation between the different systems, Is the speed of light.
3. The intelligent automobile transverse positioning method based on low-orbit satellite-enhanced GNSS and visual multi-domain fusion according to claim 1 is characterized in that the method for creating the reference high-precision map is to use map editing software to create a reference map road network based on a high-definition satellite remote sensing image or a map of a map service provider.
4. The intelligent automobile transverse positioning method based on the low-orbit satellite-enhanced GNSS and visual multi-domain fusion, which is disclosed by claim 1, is characterized in that the method for detecting the Lane line by using the visual sense is that a vehicle-mounted camera is used for acquiring a Lane image, and the Lane image is input into an Ultra-Fast-Lane-Detection-V2 deep learning algorithm for Lane line Detection.
5. The intelligent automobile transverse positioning method based on the low-orbit satellite-enhanced GNSS and visual multi-domain fusion according to claim 1, wherein in step 4.1, two edge lines of each lane line are extracted finely by using a canny edge detection and weighted least squares fitting mode.
6. The intelligent automobile transverse positioning method based on low-orbit satellite-enhanced GNSS and visual multi-domain fusion according to claim 1 is characterized in that geometric constraints in a ground plane are constructed, wherein the distance between two edge lines of the same lane line is fixed, and parallel relations exist between the edge lines, so that the geometric constraints in the ground plane are constructed.
7. The intelligent automobile transverse positioning method based on low-orbit satellite-enhanced GNSS and visual multi-domain fusion according to claim 1, wherein a conversion relation between image pixel coordinates and ground plane coordinates is established by using a rotation matrix R and a translation vector t: ; In the formula, 、 Respectively the abscissa and ordinate of the image pixels, As a scale factor of the dimensions of the device, As an internal reference matrix of the camera, Respectively the first and second column elements of the rotation matrix R, In order to translate the vector of the vector, 、 The abscissa and ordinate in the ground plane coordinate system, respectively.
8. The intelligent automobile transverse positioning method based on low-orbit satellite-enhanced GNSS and visual multi-domain fusion according to claim 1, wherein the formula for determining the plane normal vector is expressed as: ; Wherein, the Is the number of geometric constraints of the ground, For errors associated with the plane normal vector N, the geometric property calculated from a given plane normal vector N , Is a known and geometrical attribute The determined value of the correlation is set, Is a sampling plane normal vector.

Description

Intelligent automobile transverse positioning method based on low-orbit satellite enhanced GNSS and visual multi-domain fusion Technical Field The invention relates to the field of satellite positioning and computer vision, in particular to an intelligent automobile transverse positioning method based on low-orbit satellite enhanced GNSS and vision multi-domain fusion. Background Advances in science and technology are continuously pushing intelligent development of automobiles, and high-precision positioning is an important ring in intelligent development of automobiles. For intelligent automobiles, highly accurate position information is a precondition for sensing traffic conditions, planning driving paths and deciding driving behaviors. As a basis for normal operation of an automatic driving system, along with the continuous development and popularization of an automatic driving technology, the requirements for high-precision positioning of intelligent automobiles are increasing, and particularly in the aspect of transverse positioning, the transverse positioning enables the intelligent automobiles to not only determine global position information of the current path, but also acquire a specific lane of the current path and the accurate position in the lane, so that the intelligent automobiles can realize more accurate path planning and navigation. At present, intelligent automobile transverse positioning mainly depends on comprehensive application of technologies such as satellite navigation systems (such as GPS (global positioning system), beidou and the like), vehicle-mounted sensors, high-precision maps and the like. The satellite navigation system is the most convenient and quick means for the current intelligent automobile positioning, and the vehicle can determine the longitude, latitude and height information of the vehicle by receiving satellite signals, so that the positioning on a map is realized. However, since the satellite signals have canyon effect in high building dense areas, the satellite signals are easy to influence, and signal dead zones exist in areas such as tunnels, so that the positioning accuracy and range are limited. The satellite orbit height of the low orbit satellite (LEO) is lower than 2000km from the earth surface, and compared with the medium and high orbit satellite, the low orbit satellite has the characteristics of high signal intensity, rich frequency spectrum resources and the like, and can effectively enhance the signal quality and the anti-interference capability of the receiver. Therefore, the low-orbit satellite can effectively make up for the deficiency of GNSS positioning, so that high-precision transverse positioning is possible in more scenes. Meanwhile, the application of the vehicle-mounted sensor can collect environmental information around the vehicle, such as road information and the like, and provides more reference data for transverse positioning. In addition, the high-precision map is also an indispensable part of the intelligent automobile high-precision positioning, and can provide detailed information of roads, including lane lines, traffic signs, intersection structures and the like, so that the positioning precision and the driving safety are improved. At present, a single data source is difficult to meet the requirement of high-precision transverse positioning in a complex traffic scene. Disclosure of Invention In order to solve the defects in the prior art, the invention provides an intelligent automobile transverse positioning method based on the integration of a low-orbit satellite enhanced GNSS and visual multi-domain, the invention enhances the positioning navigation information of the GNSS through the low-orbit satellite, and provides a low-cost reference map creation method, and obtaining a driving path by carrying out map matching based on a weight model on a reference map, judging a specific lane of the current path by visually detecting lane lines, and finally calculating the transverse position in the current lane based on a Gao Sibei leaf model to realize high-precision intelligent automobile transverse positioning. The technical scheme adopted by the invention is as follows: An intelligent automobile transverse positioning method based on low-orbit satellite enhanced GNSS and visual multi-domain fusion comprises the following steps: Step 1, pseudo-range single-point positioning and precise single-point positioning data processing are carried out on a low-orbit enhanced GNSS according to a precise single-point positioning model of the low-orbit satellite enhanced GNSS, and longitude and latitude coordinates after enhanced positioning are obtained; Step 2, creating a reference high-precision map, carrying out map matching based on a weight model on the reference high-precision map, and screening to obtain an initial path; Step 3, obtaining an initial path based on screening, and utilizing vision to detect lane lines, and confirming a