Urban scene visual geographic positioning method and device based on sequence spherical image
Through the visual geolocation method based on sequence spherical images, combined with vocabulary trees and multiple trajectory smoothing technologies, the efficiency and accuracy of spherical panoramic image positioning in complex urban scenes is solved, and efficient and accurate visual geolocation is achieved.
Patent Information
- Application Number
- CN202411050661.5
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2024-08-01
- Publication Date
- 2024-11-12
AI Technical Summary
The existing visual geolocation methods cannot meet the visual positioning needs based on spherical panoramic images, especially in complex urban scenes, where camera continuous tracking and high data acquisition cost are difficult.
Using the visual geolocation method of urban scenes based on sequence spherical images, by obtaining panoramic database images and building visual maps, using the visual geolocation method of the vocabulary tree combined with the matching of 2D-3D corresponding points, and filtering through linear interpolation, velocity constraints, and sliding windows Methods such as Kalman filtering perform trajectory smoothing to improve positioning accuracy and efficiency.
It realizes efficient and precise visual geolocation based on spherical panoramic images, improves positioning efficiency and accuracy, and solves the problems of difficulty in camera tracking and high data acquisition costs in traditional methods.
Smart Images

Figure CN118936452A_ABST
Abstract
Description
Technical Field
[0001] The present application relates to the field of photogrammetry and computer vision, and in particular to a method and device for visual geographic positioning of urban scenes based on sequential spherical images. Background Art
[0002] The demand for precise positioning in urban scenes is becoming increasingly urgent, and visual geolocation technology has quickly become the focus due to its excellent accuracy. This technology combines advanced image processing, computer vision and geographic information systems to provide highly accurate positioning and navigation services in complex urban environments. Traditional visual geolocation technology is usually based on perspective cameras, but perspective cameras have problems such as difficulty in continuous camera tracking and high data collection costs for complex urban scenes.
[0003] Spherical panoramic cameras, also known as 360 cameras, can cover a 360°×180° viewing angle range of the collection point and obtain images with a full range of viewing angles. They have been widely used in security monitoring, street view maps, robot navigation and other fields. Compared with flat perspective cameras, spherical panoramic cameras have a larger field of view, can achieve continuous viewing angle tracking in complex scenes, and reduce the workload of data collection, so they have great potential in the positioning of complex urban scenes.
[0004] However, storing spherical images in the equirectangular projection (ERP) format will cause distortion. Due to the different shooting directions of the camera, the images captured at the same location may be very different, which will affect the positioning. Existing research has few solutions for visual geolocation using spherical images, or focuses on the improvement of a single step, lacking a comprehensive visual positioning solution for spherical panoramic images. With the popularization of spherical panoramic cameras, such as consumer-grade Insta360, Ricoh THETA, etc., and the development of image processing technology, the demand for visual geography based on spherical panoramic images will become increasingly greater.
[0005] In summary, the existing technical solutions cannot meet the visual positioning requirements based on spherical panoramic images. Summary of the invention
[0006] The purpose of the present invention is to provide a method and device for visual geographic positioning of urban scenes based on sequential spherical images in order to solve the problem that the existing visual geographic positioning methods cannot meet the visual positioning requirements based on spherical panoramic images.
[0007] The above-mentioned purpose of the present application is achieved through the following technical solutions: S1: Obtain panoramic database images and build visual maps; S2: Get the image to be located; S3: A visual geolocation method based on a vocabulary tree combines panoramic database images and visual maps to obtain 2D-3D corresponding points between the image to be located and the visual map; S4: Determine the positioning result through the 2D-3D corresponding points between the image to be positioned and the visual map; S5: Smooth the trajectory of the positioning result to achieve visual geolocation.
[0008] Optionally, step S1 includes: S11: Obtain panoramic database images through a mobile measurement system; S12: Perform feature matching on the panoramic database image to obtain matching results, including: image feature extraction based on SIFT algorithm, feature matching taking into account spatial and sequential constraints, and essential matrix Gross error elimination under constraints; S13: Perform 3D point triangulation based on the posture information and matching results of the panoramic database image to construct a sparse point cloud; S14: Perform global BA optimization on the sparse point cloud to generate a visual map.
[0009] Optionally, step S12 includes: S12a: Use the GPU version of the SIFT algorithm, SIFTGPU algorithm, to extract image features and generate feature descriptors for the panoramic database images; S12b: Based on the location information of the panoramic database images, spatial and order constraints are used to guide the selection of matching pairs; S12c: Essential matrix through RANSAC algorithm Constrained estimation,eliminates outliers introduced by feature descriptors.
[0010] Optionally, step S12c includes: Use vector to plane geodesic angle error metric; Spherical angle error Using formula (4) (4) in Indicate point The transposed vector of ; Indicate point A vector of and These are the spherical coordinates of corresponding points in the image; Functions that represent absolute values; Pixel error threshold Convert to spherical angle ,as follows: (5) in, The ratio factor between two units of measurement. Indicates the width of the image; Indicates the height of the image; represents the maximum value function; It is the gross error judgment threshold measured by angle; pixel error threshold Set to 4 pixels; like , then the corresponding matching point is marked as an internal point. The corresponding points with the same name are marked as outliers, and the outliers are outliers.
[0011] Optionally, step S3 includes: S31: According to the visual geographic positioning method based on the vocabulary tree, combined with the image to be positioned, the corresponding panoramic database image is matched, as follows: Use SIFTGPU algorithm to extract feature points of the image to be located; Through the image feature index, the panoramic database images similar to the image to be located are retrieved and matched; S32: Obtaining 2D-3D corresponding points between the matched panoramic database image and the visual map; S33: Based on the PnP algorithm, the 2D-3D corresponding points between the panoramic database image and the visual map are combined to obtain the 2D-3D corresponding points between the image to be located and the visual map.
[0012] Optionally, step S4 further includes: S41: for the image to be positioned that fails to be positioned, use adjacent positioning image frames to estimate the position of the image to be positioned that fails to be positioned; S42: based on the estimated position, re-retrieve from the panoramic database image to obtain spatially adjacent candidate images; S43: Re-positioning is performed based on spatially adjacent candidate images.
[0013] Optionally, step S5 includes: S51: Perform linear interpolation position estimation on the positioning result: Given time Location and time Location , estimate the time by linear interpolation Location ; (7) in and ;Location That is, in time The positioning result at the time S52: Based on the positioning results, the speed of the vehicle at different times when the image is collected is calculated, and when the vehicle speed is close to zero, the continuous stationary positioning results are fitted through speed constraints; S53: Perform sliding window filtering on the trajectory data of the positioning result; S54: Perform Kalman filtering on the trajectory data of the positioning result. The specific steps are as follows: S54a: Construct a dynamic model of the spherical camera to describe the motion state; the motion state includes: position motion state and speed parameters; set the distance threshold between the predicted result and the actual positioning result to ; S54b: Based on the current moment Corresponding position predict Vehicle position at the moment ; S54c: Calculation Positioning result at the moment And the prediction results The distance between ; S54d: If Then mark the current point as an outlier and filter it, and repeat steps S54b to S54d; if The prediction result is taken as The actual positioning result at the moment is used to update the position and obtain the final positioning result.
[0014] An electronic device includes a processor, a memory, a user interface and a network interface, wherein the memory is used to store instructions, the user interface and the network interface are used to communicate with other devices, and the processor is used to execute the instructions stored in the memory so that the electronic device executes a method for visual geographic positioning of urban scenes based on sequential spherical images.
[0015] A computer-readable storage medium stores instructions, and when the instructions are executed, a method for visual geographic positioning of urban scenes based on sequential spherical images is executed.
[0016] The beneficial effects of the technical solution provided by this application are: 1. A visual geolocation method based on vocabulary tree was designed to accelerate the image matching process and improve positioning efficiency and accuracy.
[0017] 2. Aiming at the visual geographic positioning goal of sequential panoramic images, a positioning-based retrieval optimization method is designed. For the images to be positioned that fail to be positioned, the position of the images to be positioned that fail to be positioned is estimated using adjacent positioning image frames; based on the estimated position, the images are re-retrieved from the panoramic database images to obtain spatially adjacent candidate images; and the positioning is re-performed based on the spatially adjacent candidate images.
[0018] 3. The positioning results are adjusted by using a trajectory smoothing method that integrates linear interpolation position estimation, stationary point velocity constraint, sliding window filtering, and Kalman filtering to improve positioning accuracy. BRIEF DESCRIPTION OF THE DRAWINGS
[0019] The present application will be further described below with reference to the accompanying drawings and embodiments, in which: Figure 1 is a step diagram in an embodiment of the present application; Figure 2 It is a projection model diagram in the embodiment of the present application; Figure 3 is a visual map in an embodiment of the present application; Figure 4 is a flowchart of search optimization positioning in an embodiment of the present application; Figure 5 It is a schematic diagram of the structure of an electronic device in an embodiment of the present application. DETAILED DESCRIPTION
[0020] In order to have a clearer understanding of the technical features, purposes and effects of the present application, the specific implementation methods of the present application are now described in detail with reference to the accompanying drawings.
[0021] The embodiments of the present application provide a method for visual geographic positioning of urban scenes based on sequential spherical images.
[0022] Please refer to Figure 1 , Figure 1 This is a step diagram of a method for visual geographic positioning of urban scenes based on sequential spherical images in an embodiment of the present application, comprising: S1: Obtain panoramic database images and build visual maps; S2: Get the image to be located; S3: A visual geolocation method based on a vocabulary tree combines panoramic database images and visual maps to obtain 2D-3D corresponding points between the image to be located and the visual map; Specifically, the main principle of visual geolocation is to match the image to be located with the database image used to construct the visual map, obtain enough 2D-3D corresponding points and use the PnP (Perspective-n-Point) algorithm to achieve positioning.
[0023] S4: Determine the positioning result through the 2D-3D corresponding points between the image to be positioned and the visual map; S5: Smooth the trajectory of the positioning result to achieve visual geolocation.
[0024] Step S1 includes: S11: Obtain panoramic database images through a mobile measurement system; Specifically, during the reconstruction process, the first step is to read the parameter file to obtain the internal parameter settings of the camera. Figure 2 The projection model shown, the center coordinates of the spherical image are set to ( , ), the radius of the spherical model of the projected image is usually set to 1. Regarding the camera external parameters, the present invention uses a mobile measurement system MMS (Mobile Mapping System) equipped with a multi-frequency GNSS (Global Navigation Satellite System) receiver and an IMU (Inertial Measurement Unit) to obtain each image and its pose information, and ensures the accuracy of the initial pose by monitoring the pose and position in real time.
[0025] S12: Perform feature matching on the panoramic database image to obtain matching results, including: image feature extraction based on SIFT algorithm, feature matching taking into account spatial and sequential constraints, and essential matrix Gross error elimination under constraints; S13: Perform 3D point triangulation based on the posture information and matching results of the panoramic database image to construct a sparse point cloud; Specifically, it traverses each image with a successfully initialized pose and finds feature points in other images that correspond to the successfully matched feature points in the current image. When the number of matches exceeds a certain threshold, the feature points on the image can be mapped to the point cloud coordinate system based on the principle that the common viewpoint and the coordinate origin in the spherical coordinate system of the two images form three collinear points.
[0026] S14: Perform global BA optimization on the sparse point cloud to generate a visual map.
[0027] Specifically, the errors between all observed image points and their corresponding 3D points are first expressed as residuals. Then, an optimization algorithm is used to iteratively adjust the camera parameters and the positions of the 3D points to minimize the residual sum of squares (as shown in formula (6)).
[0028] (6) Finally, the point cloud data is cleaned by applying multiple filtering criteria to the 3D reconstructed points. Points that do not meet the criteria are filtered out based on the maximum reprojection error and the minimum triangulation angle. The remaining points are then checked individually, and if the number of matches for a point in the multi-view image is less than the specified minimum, or if the height of the point is outside the specified range, the point is removed. Once the above steps are completed, a point cloud is generated. Figure 3 Visual map shown.
[0029] Specifically, global BA (Bundle Adjustment) optimization.
[0030] Step S12 includes: S12a: Use the GPU version of the SIFT algorithm, SIFTGPU algorithm, to extract image features and generate feature descriptors for the panoramic database images; Specifically, image feature extraction based on the SIFT algorithm: Although ERP projection introduces geometric distortion, the present invention still uses the classic SIFT (Scale Invariant Feature Transform) algorithm for feature extraction. On the one hand, most data acquisition fixes the roll angle and pitch angle of the camera, such as the ground mobile measurement vehicle acquisition system, to ensure the consistency of the ERP image geometric structure near the spherical equator; on the other hand, the classic SIFT algorithm has good scale and rotation invariance, strong resistance to viewing angle changes, and has a corresponding high-performance open source library. Specifically, the present invention uses the GPU (Graphic Processing Unit) version of the SIFT algorithm, SIFTGPU, to extract image features and generate descriptors.
[0031] S12b: Based on the location information of the panoramic database images, spatial and order constraints are used to guide the selection of matching pairs; Specifically, feature matching taking into account spatial and sequential constraints: Based on the position information of the panoramic database image, the present invention uses spatial and sequential constraints to guide the selection of matching pairs to prevent wasting computing resources and introducing false matches. The SIFT GPU library is used to accelerate feature matching, thereby achieving efficient and high-precision feature matching.
[0032] S12c: Essential matrix through RANSAC algorithm Constrained estimation,eliminates outliers introduced by feature descriptors.
[0033] Step S12c includes: Use vector to plane geodesic angle error metric; Spherical angle error Using formula (4) (4) in Indicate point The transposed vector of ; Indicate point A vector of and These are the spherical coordinates of corresponding points in the image; Functions that represent absolute values; Pixel error threshold Convert to spherical angle ,as follows: (5) in, The ratio factor between two units of measurement. Indicates the width of the image; Indicates the height of the image; represents the maximum value function; It is the gross error judgment threshold measured by angle; pixel error threshold Set to 4 pixels; like , then the corresponding matching point is marked as an internal point. The corresponding points with the same name are marked as outliers, and the outliers are outliers.
[0034] Specifically, the essential matrix Constrained gross error elimination: The SIFT algorithm generates feature descriptors by calculating local image patches around key points, which may introduce outliers in the initial matching process and lead to incorrect matching results. To solve this problem, the present invention adopts a hypothesis verification framework based on RANSAC (Random sample consensus) for robust estimation, which relies on the error metric and gross error judgment threshold Matching points, also called corresponding points or points of the same name, refer to the same points of the same object identified in different perspectives or different images.
[0035] Step S3 includes: S31: According to the visual geographic positioning method based on the vocabulary tree, combined with the image to be positioned, the corresponding panoramic database image is matched, as follows: Use SIFTGPU algorithm to extract feature points of the image to be located; Through the image feature index, the panoramic database images similar to the image to be located are retrieved and matched; S32: Obtaining 2D-3D corresponding points between the matched panoramic database image and the visual map; In terms of feature matching, exhaustive matching consumes a lot of time and resources. In order to achieve efficient feature matching, the present invention chooses to use a vocabulary tree to accelerate the matching process. Specifically, the present invention first constructs an image feature index based on the panoramic database image used to establish a visual map. For the image frame that needs to be located, the candidate matching image is quickly found according to the index, and the SIFT GPU library is used for feature matching.
[0036] S33: Based on the PnP algorithm, the 2D-3D corresponding points between the panoramic database image and the visual map are combined to obtain the 2D-3D corresponding points between the image to be located and the visual map.
[0037] Specifically, after completing the matching, the next step is to accurately geolocate (absolute orientation) the image to be positioned. In perspective imaging, the PnP algorithm can be used to directly calculate the absolute pose of the newly added image based on the 2D-3D correspondence relationship. Since it still satisfies the collinear relationship between the projection center, the image point, and the scene point, PnP can also be applied to the absolute orientation of spherical images. Based on the 2D-3D corresponding points between the panoramic database image and the visual map, the 2D-3D corresponding points between the query image and the visual map can be obtained. In addition, the present invention uses the RANSAC algorithm to perform robust estimation of outliers based on the angle error metric.
[0038] Step S4 also includes: S41: for the image to be positioned that fails to be positioned, use adjacent positioning image frames to estimate the position of the image to be positioned that fails to be positioned; S42: based on the estimated position, re-retrieve from the panoramic database image to obtain spatially adjacent candidate images; S43: Re-positioning is performed based on spatially adjacent candidate images.
[0039] Specifically, the actual positions of the sequence spherical images are continuous. For image frames that cannot be located due to matching failure caused by erroneous retrieval, the present invention uses adjacent positioning image frames to estimate their positions. Figure 4 As shown, for images that fail to be positioned, the present invention first searches for image frames near them in the time series, assuming that the posture data of these adjacent frames have been successfully obtained and are sufficiently accurate. Then, by interpolating or fitting the position data of these adjacent frames, the present invention estimates the possible position of the image that failed to be positioned. Based on the estimated position, the present invention re-retrieves candidate images that are spatially adjacent to it from the database, and then re-matches and re-positions it. In addition, in the actual matching process, the present invention dynamically adjusts the minimum number of inliers in the positioning algorithm, and by iteratively performing positioning-based retrieval optimization, a high success rate positioning of sequential panoramic images can be achieved.
[0040] Specifically, when spherical images are stored in ERP format, distortion is inevitable. In the actual positioning process, since the perspective of the panoramic database image collected by the mobile measurement system MMS cannot be completely aligned with the perspective of the image to be positioned, differences may occur even in spherical images captured at the same location. This results in the inability to retrieve the correct candidate matching image through the vocabulary tree during the retrieval process. In order to solve this problem, the present invention proposes a sequential spherical image retrieval optimization method based on known positioning results.
[0041] Step S5 includes: S51: Perform linear interpolation position estimation on the positioning result: Given time Location and time Location , estimate the time by linear interpolation Location ; (7) in and ;Location That is, in time The positioning result at the time Specifically, after positioning, some images may have obvious distortion or occlusion problems. Even if they are matched with the correct image, positioning may fail due to insufficient 2D-3D correspondence points. When the time interval between the sequence images is small enough and the vehicle does not change direction suddenly during the data collection process, the position of the nearby images can be used to estimate its position.
[0042] S52: Based on the positioning results, the speed of the vehicle at different times when the image is collected is calculated, and when the vehicle speed is close to zero, the continuous stationary positioning results are fitted through speed constraints; Specifically, the stationary point speed constraint: The present invention introduces speed constraint technology to solve the positioning fluctuation problem that occurs when the vehicle is stationary. Specifically, the speed of the vehicle at different times when the image is collected is first calculated based on the positioning results. When the system detects that the vehicle speed is close to zero, the speed constraint will be applied to fit the continuous stationary positioning results to prevent the positioning results from fluctuating when the vehicle stops.
[0043] S53: Perform sliding window filtering on the trajectory data of the positioning result; Specifically, by averaging the trajectory data over a certain period of time, the sliding window filter can effectively reduce short-term random noise and errors. In the present invention, the sliding window size is usually 5. In the face of a trajectory segment with excessive noise, the sliding window filter may not be able to eliminate the fluctuations in the trajectory. Therefore, the present invention further introduces the Kalman filter technology.
[0044] S54: Perform Kalman filtering on the trajectory data of the positioning result. The specific steps are as follows: S54a: Construct a dynamic model of the spherical camera to describe the motion state; the motion state includes: position motion state and speed parameters; set the distance threshold between the predicted result and the actual positioning result to ; S54b: Based on the current moment Corresponding position predict Vehicle position at the moment ; S54c: Calculation Positioning result at the moment And the prediction results The distance between ; S54d: If Then mark the current point as an outlier and filter it, and repeat steps S54b to S54d; if The prediction result is taken as The actual positioning result at the moment is used to update the position and obtain the final positioning result.
[0045] Specifically, in the process of visual geolocation, some applications have clear requirements for the continuity and smoothness of the positioning trajectory. For example, in the underground pipeline measurement task, any discontinuity or sudden change in the trajectory may cause positioning errors and affect the subsequent pipeline maintenance and management tasks. Therefore, it is necessary to adopt appropriate strategies to optimize the positioning trajectory to meet the specific requirements of the application. The present invention makes full use of the position continuity of the sequence panoramic image and adopts multiple methods to achieve trajectory smoothing.
[0046] The present application also discloses an electronic device. Figure 5 , Figure 5 The electronic device 500 may include: at least one processor 501 , at least one network interface 504 , a user interface 503 , a memory 505 , and at least one communication bus 502 .
[0047] The communication bus 502 is used to realize the connection and communication between these components.
[0048] The user interface 503 may include a display screen, and the optional user interface 503 may also include a standard wired interface or a wireless interface.
[0049] The network interface 504 may optionally include a standard wired interface or a wireless interface (such as a WI-FI interface).
[0050] The present application also discloses a computer-readable storage medium storing a plurality of instructions suitable for loading by a processor to execute the above-mentioned urban scene visual geographic positioning method based on sequential spherical images.
[0051] The above are only exemplary embodiments of the present disclosure and cannot be used to limit the scope of the present disclosure. That is, any equivalent changes and modifications made according to the teachings of the present disclosure are still within the scope of the present disclosure. After considering the disclosure of the specification and the truth of practice, those skilled in the art will easily think of other embodiments of the present disclosure.
[0052] This application is intended to cover any variation, use or adaptation of the present disclosure, which follows the general principles of the present disclosure and includes common knowledge or customary technical means in the art not described in the present disclosure. The description and examples are to be regarded as exemplary only, and the scope and spirit of the present disclosure are defined by the claims.
Claims
1. A method for visual geographic positioning of urban scenes based on sequential spherical images, characterized in that: The method comprises the following steps: S1: Obtain panoramic database images and build visual maps; S2: Get the image to be located; S3: A visual geolocation method based on a vocabulary tree combines panoramic database images and visual maps to obtain 2D-3D corresponding points between the image to be located and the visual map; S4: Determine the positioning result through the 2D-3D corresponding points between the image to be positioned and the visual map; S5: Smooth the trajectory of the positioning result to achieve visual geolocation.
2. The urban scene visual geographic positioning method based on sequential spherical images according to claim 1, characterized in that: Step S1 includes: S11: Obtain panoramic database images through a mobile measurement system; S12: Perform feature matching on the panoramic database image to obtain matching results, including: image feature extraction based on SIFT algorithm, feature matching taking into account spatial and sequential constraints, and essential matrix Gross error elimination under constraints; S13: Perform 3D point triangulation based on the posture information and matching results of the panoramic database image to construct a sparse point cloud; S14: Perform global BA optimization on the sparse point cloud to generate a visual map.
3. The urban scene visual geographic positioning method based on sequential spherical images as claimed in claim 2, characterized in that: Step S12 includes: S12a: Use the GPU version of the SIFT algorithm, SIFTGPU algorithm, to extract image features and generate feature descriptors for the panoramic database images; S12b: Based on the location information of the panoramic database images, spatial and order constraints are used to guide the selection of matching pairs; S12c: Essential matrix through RANSAC algorithm Constrained estimation,eliminates outliers introduced by feature descriptors.
4. The urban scene visual geographic positioning method based on sequential spherical images as claimed in claim 3, characterized in that: Step S12c includes: Use vector to plane geodesic angle error metric; Spherical angle error Using formula (4) (4) in Indicate point The transposed vector of ; Indicate point A vector of and These are the spherical coordinates of corresponding points in the image; Functions that represent absolute values; Pixel error threshold Convert to spherical angle ,as follows: (5) in, represents the ratio of two units of measurement, Indicates the width of the image; Indicates the height of the image; represents the maximum value function; It is the gross error judgment threshold measured by angle; pixel error threshold Set to 4 pixels; like , then the corresponding matching point is marked as an internal point. The corresponding points with the same name are marked as outliers, and the outliers are outliers.
5. The urban scene visual geographic positioning method based on sequential spherical images according to claim 1, characterized in that: Step S3 includes: S31: According to the visual geographic positioning method based on the vocabulary tree, combined with the image to be positioned, the corresponding panoramic database image is matched, as follows: Use SIFTGPU algorithm to extract feature points of the image to be located; Through the image feature index, the panoramic database images similar to the image to be located are retrieved and matched; S32: Obtaining 2D-3D corresponding points between the matched panoramic database image and the visual map; S33: Based on the PnP algorithm, the 2D-3D corresponding points between the panoramic database image and the visual map are combined to obtain the 2D-3D corresponding points between the image to be located and the visual map.
6. The urban scene visual geographic positioning method based on sequential spherical images according to claim 1, characterized in that: Step S4 also includes: S41: for the image to be positioned that fails to be positioned, use adjacent positioning image frames to estimate the position of the image to be positioned that fails to be positioned; S42: based on the estimated position, re-retrieve from the panoramic database image to obtain spatially adjacent candidate images; S43: Re-positioning is performed based on spatially adjacent candidate images.
7. The urban scene visual geographic positioning method based on sequential spherical images according to claim 1, characterized in that: Step S5 includes: S51: Perform linear interpolation position estimation on the positioning result: Given time Location and time Location , estimate the time by linear interpolation Location ; (7) in and ;Location That is, in time The positioning result at the time S52: Based on the positioning results, the speed of the vehicle at different times when the image is collected is calculated, and when the vehicle speed is close to zero, the continuous stationary positioning results are fitted through speed constraints; S53: Perform sliding window filtering on the trajectory data of the positioning result; S54: Perform Kalman filtering on the trajectory data of the positioning result. The specific steps are as follows: S54a: Construct a dynamic model of the spherical camera to describe the motion state; the motion state includes: position motion state and speed parameters; set the distance threshold between the predicted result and the actual positioning result to ; S54b: Based on the current moment Corresponding position predict Vehicle position at the moment ; S54c: Calculation Positioning result at the moment And the prediction results The distance between ; S54d: If Then mark the current point as an outlier and filter it, and repeat steps S54b to S54d; if The prediction result is taken as The actual positioning result at the moment is used to update the position and obtain the final positioning result.
8. An electronic device, characterized in that: The electronic device comprises a processor (501), a memory (505), a user interface (503) and a network interface (504), wherein the memory (505) is used to store instructions, the user interface (503) and the network interface (504) are used to communicate with other devices, and the processor (501) is used to execute the instructions stored in the memory (505) so that the electronic device executes the method according to any one of claims 1 to 7.
9. A computer-readable storage medium, characterized in that: The computer-readable storage medium stores instructions, and when the instructions are executed by a computer, the method according to any one of claims 1 to 7 is executed.
