GNSS (Global Navigation Satellite System) / visual loose coupling positioning method fusing environmental semantics

Abstract

The invention belongs to the technical field of navigation and positioning, particularly relates to a GNSS / visual loose coupling positioning method fusing environmental semantics, and solves the technical problem that a traditional method is insufficient in positioning precision when facing an urban sheltered environment and a high-dynamic scene. According to the method, a GNSS receiver obtains position coordinates of a carrier through a double-difference measurement model of the GNSS receiver, and continuous image frames captured by a stereo camera are transmitted to a VO module for executing pose estimation of the carrier; carrier position coordinates of a GNSS and carrier pose information of a VO module are combined by adopting a loose coupling model, then a shielding factor is calculated according to environmental semantic information and a local mapping module, an adaptive KF algorithm is introduced in combination with the shielding factor, and under the constraint of the shielding factor, the adaptive KF algorithm can calculate an optimal pose estimation value, so that the optimal pose estimation value is obtained. The value is fed back to the GNSS to correct the position coordinate of the carrier, and finally the corrected position coordinate is output as a positioning result.

Description

Technical Field

[0001] The present invention relates to the technical field of navigation and positioning, and in particular to a GNSS / vision loose coupling positioning method integrating environmental semantics. Background Art

[0002] With the development of the intelligent transportation field, the positioning and navigation technology of autonomous vehicles has become a current research hotspot. Reliable, continuous, and high-precision positioning technology is the key to realizing autonomous driving. The Global Navigation Satellite System (GNSS) can provide reliable positioning services in an open sky environment. However, in complex environments such as urban canyons or under tree shades, GNSS signals are restricted, and their reliability and continuity will deteriorate significantly. Integrating GNSS with other positioning sensors to achieve combined positioning is an effective way to improve the positioning performance of GNSS signals in restricted environments. The vision positioning system is a low-cost and highly reliable navigation system that can provide navigation information in environments where GNSS signals are restricted.

[0003] Visual Odometry (VO) uses visual sensors to capture environmental information and track the camera movement in real time to achieve precise positioning of the target. The VO module realizes positioning based on the assumption that environmental objects are stationary, and the dynamic objects commonly present in urban streets will cause incorrect matching of feature points, affecting the accuracy of pose estimation. Summary of the Invention

[0004] To solve the technical defect of insufficient positioning accuracy of traditional methods in the face of urban occlusion environments and high-dynamic scenarios, the present invention provides a GNSS / vision loose coupling positioning method integrating environmental semantics.

[0005] The present invention provides a GNSS / vision loose coupling positioning method, which is implemented based on a stereo camera, GNSS, and a SLAM system. The stereo camera is fixedly connected to the carrier to collect images during the carrier's driving in real time. The SLAM system includes a VO module and a local mapping module. The VO module includes a feature point tracking unit. The steps of the loose coupling positioning method are as follows:

[0006] Step 1: Improve the VO module by embedding a real-time environmental semantic segmentation processing unit into the VO module. The environmental semantic segmentation processing unit works in parallel with the feature point tracking unit. Images are input into the VO module in the form of consecutive frames. The feature point tracking unit extracts ORB (Oriented FAST and Rotated BRIEF) feature points from the images to form a feature map containing dynamic and static feature points. The environmental semantic segmentation processing unit performs environmental semantic segmentation on the images to generate a masked image with specific semantic labels. The VO module fuses the information of the feature map and the masked image, marks the dynamic feature points as outliers, and removes the outliers to obtain an image with only static feature points.

[0007] Step 2: The VO module performs point-by-point matching of the static feature points of the current image with those of the previous frame, and then obtains the six-degree-of-freedom rotation R and translation T of the stereo camera by minimizing the reprojection error.

[0008] Step 3: Independently obtain the preliminary position coordinates of the carrier based on the GNSS double-difference observation model.

[0009] Step 4: The rotation R and translation T of the stereo camera are the rotation R and translation T of the carrier. Use the rotation R and translation T of the carrier to construct a visual positioning error model, where the visual positioning error model is:

[0010] (1),

[0011] In formula (1), 、 and are the length scale factor of the stereo camera, the position coordinates of the carrier, and the attitude of the carrier in the navigation coordinate system respectively. Among them, the position coordinates of the carrier are obtained by integrating the translation T of the carrier, and the attitude of the carrier is obtained by integrating the rotation R of the carrier; 、 and are the errors of the length scale factor, position coordinates, and attitude in the navigation coordinate system respectively; 、 and are the continuous time derivatives of the errors of the length scale factor, position coordinates, and attitude in the navigation coordinate system respectively; is the estimated rotation matrix from the carrier coordinate system to the navigation coordinate system; is the estimated translation in the carrier coordinate system; is the estimated value of the length scale factor; and are the translation error and rotation error in the carrier coordinate system respectively; is Random drift;

[0012] Simplify formula (1) into a system model, which is expressed as:

[0013] (2),

[0014] In formula (2), is ; is ; is ; is ; is ;

[0015] Then, construct the state model form of the KF algorithm through formula (2):

[0016] (3),

[0017] In formula (3), is the state vector at epoch , is the discrete-time transition matrix from epoch to epoch , is the noise distribution matrix, is the system noise sequence;

[0018] Adopt a loose coupling model, and use the difference between the carrier position coordinates obtained by the VO module and the preliminary carrier position coordinates obtained by GNSS in step 3 as the observation value input of the KF algorithm, and construct the observation model of the KF algorithm as:

[0019] (4),

[0020] In formula (4), and are the carrier position coordinates obtained by the VO module and GNSS respectively; is the observation transition matrix; is the observation noise;

[0021] Then, the state update equation of the KF algorithm is expressed as:

[0022] (5),

[0023] In formula (5), represents the error covariance at time epoch, represents the corresponding covariance matrix; Indicates an epoch To the epoch The prior state vector estimate value; Indicates from the epoch To the epoch The state transition matrix; Indicates an epoch The state vector estimate value; Indicates an epoch To the epoch The one-step covariance of the predicted estimation error; Indicates an epoch The predicted estimation error covariance; Indicates from the epoch To the epoch The transpose of the state transition matrix; Indicates the transpose of the noise distribution matrix;

[0024] Where the observation update equation of the KF algorithm is expressed as:

[0025] (6),

[0026] In formula (6), Indicates the gain matrix; Indicates the transpose of the observation transfer matrix; Indicates the observation noise covariance; Indicates an epoch The state vector estimate value; Indicates an epoch The predicted estimation error covariance;

[0027] Through the observation update equation of the KF algorithm, the position coordinates of the carrier can be updated, and the updated position coordinates of the carrier are fed back to the GNSS. The GNSS corrects the preliminary position coordinates of the carrier in step 3 through the updated position coordinates of the carrier, and finally the corrected position coordinates of the carrier are output as the positioning result.

[0028] Preferably, in step 4, after obtaining the observation update equation of the KF algorithm, it further includes an adaptive adjustment of the observation noise covariance ; The masked image with specific semantic labels obtained in step 1 contains an occluder semantic label. The depth information of the occluder is obtained through the local mapping module of the SLAM system. The depth information of the occluder is the distance between the carrier and the occluder, and an occlusion factor Is:

[0029] (7),

[0030] In formula (7), is the depth of field, the depth of field ranges from ;

[0031] When an occluder is detected in the image, the occlusion factor is integrated into the original observation noise covariance to obtain the observation noise covariance after adaptive adjustment , the observation noise covariance The adaptive adjustment formula is:

[0032] (8),

[0033] When an occluder is detected, the observation noise covariance after adaptive adjustment is substituted into the observation update equation of the KF algorithm, and the optimal pose estimation value of the carrier can be calculated.

[0034] The gain matrix is affected by equations (3) to (6). Specifically, when the set observation noise variance exceeds the actual noise distribution level, the corresponding value will decrease, resulting in an improper reduction in the uncertainty range of the true value of the state estimation, and thus causing a deviation in the estimation result. On the contrary, if the set is too small, the value will increase, which may cause the phenomenon of filter divergence. If there are abnormal occluders in the observation data and the system still uses the original observation noise covariance without timely adjustment, then the influence of the occluder on the filter will not be suppressed, resulting in an unclear filter convergence effect or even filter divergence.

[0035] Therefore, the adaptive adjustment of the original observation noise covariance is to improve the estimation characteristics and robustness of the KF algorithm. Integrating the occlusion factor into the original observation noise covariance and adjusting upward to obtain the observation noise covariance after adaptive adjustment , thereby reducing the gain matrix and reducing the influence of abnormal measurement data on the KF algorithm.

[0036] The technical solution provided by the present invention has the following technical effects compared with the prior art: The method of the present invention combines GNSS and visual positioning technologies. The absolute position provided by GNSS can correct the cumulative error of visual positioning, and at the same time, the local positioning ability of visual positioning can improve the positioning accuracy and stability of GNSS, effectively solving the problem of insufficient positioning accuracy of traditional methods in the face of urban occlusion environments and high-dynamic scenarios. BRIEF DESCRIPTION OF THE DRAWINGS

[0037] The accompanying drawings herein are incorporated into the specification and constitute a part of this specification, showing embodiments consistent with the present invention and, together with the specification, are used to explain the principles of the present invention.

[0038] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the following will briefly introduce the accompanying drawings required for use in the description of the embodiments or the prior art. Obviously, for those of ordinary skill in the art, without creative efforts, other drawings can also be obtained based on these drawings.

[0039] Figure 1 It is a flowchart of the GNSS / visual loose-coupling positioning method integrating environmental semantics in an embodiment of the present invention. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0040] In order to more clearly understand the above objects, features, and advantages of the present invention, the following will further describe the solution of the present invention. It should be noted that, without conflict, the embodiments of the present invention and the features in the embodiments can be combined with each other.

[0041] Many specific details are set forth in the following description to fully understand the present invention, but the present invention can also be implemented in other ways different from those described herein; obviously, the embodiments in the specification are only a part of the embodiments of the present invention, rather than all the embodiments.

[0042] The following will detail the specific embodiments of the present invention with reference to the accompanying drawings.

[0043] In one embodiment, as Figure 1 shown, a GNSS / visual loose-coupling positioning method integrating environmental semantics is disclosed. The loose-coupling positioning method is implemented based on a stereo camera, GNSS, and a SLAM (Simultaneous Localization and Mapping) system. The stereo camera is fixedly connected to the carrier to collect images during the driving of the carrier in real time, ensuring the continuity and stability of the image data. The SLAM system includes a VO module and a local mapping module. The VO module includes a feature point tracking unit. The steps of the loose-coupling positioning method are as follows:

[0044] Step 1: Improve the VO module by embedding a real-time environmental semantic segmentation processing unit into the VO module. The environmental semantic segmentation processing unit works in parallel with the feature point tracking unit. Images are input into the VO module in the form of consecutive frames. The feature point tracking unit extracts ORB feature points from the images to form a feature map containing dynamic feature points and static feature points. The environmental semantic segmentation processing unit performs environmental semantic segmentation on the images to generate a mask image with specific semantic labels. The VO module fuses the information of the feature map and the mask image, marks the dynamic feature points as outliers. The dynamic feature points are the feature points related to dynamic objects, where the dynamic objects include vehicles, pedestrians, or animals, etc., and removes the outliers to obtain an image with only static feature points. Identifying and removing the feature points related to dynamic objects can ensure that the accuracy of pose estimation is not affected by dynamic objects;

[0045] Step 2: The VO module performs point-by-point matching between the static feature points of the current image and the static feature points of the previous frame, and then obtains the six-degree-of-freedom rotation R and translation T of the stereo camera by minimizing the reprojection error;

[0046] Step 3: Independently obtain the preliminary position coordinates of the carrier based on the GNSS double-difference observation model;

[0047] Step 4: The rotation R and translation T of the stereo camera are the rotation R and translation T of the carrier. Use the rotation R and translation T of the carrier to construct a visual positioning error model, where the visual positioning error model is:

[0048] (1),

[0049] In formula (1), 、 and are the length scale factor of the stereo camera, the position coordinates of the carrier, and the attitude of the carrier in the navigation coordinate system respectively. Among them, the position coordinates of the carrier is obtained by integrating the translation T of the carrier, and the attitude of the carrier is obtained by integrating the rotation R of the carrier; 、 and are the errors of the length scale factor, position coordinates, and attitude in the navigation coordinate system respectively; 、 and are the continuous time derivatives of the errors of the length scale factor, position coordinates, and attitude in the navigation coordinate system respectively; is the estimated rotation matrix from the carrier coordinate system to the navigation coordinate system; is the estimated translation in the carrier coordinate system; is the estimated value of the length scale factor; and are the translation error and rotation error in the vehicle coordinate system, respectively; is the random drift of;

[0050] Simplify Equation (1) into a system model, and the system model is expressed as:

[0051] (2),

[0052] In Equation (2), is ; is ; is ; is ; is ;

[0053] Then, construct the state model form of the Kalman Filter (KF) algorithm through Equation (2):

[0054] X(k) = Φ(k,k - 1)X(k - 1) + T(k - 1)W(k - 1) (3),

[0055] In Equation (3), is the state vector at epoch ; is the discrete-time transition matrix from epoch to epoch ; is the noise distribution matrix, is the system noise sequence;

[0056] Adopt a loose coupling model, and use the difference between the vehicle position coordinates obtained by the VO module and the vehicle position coordinates obtained by GNSS in Step 3 as the observation value input of the KF algorithm. The observation model of the KF algorithm is constructed as:

[0057] (4),

[0058] In Equation (4), and are the vehicle positions obtained by the VO module and GNSS, respectively; is the observation transition matrix; is the observation noise;

[0059] Then, the state update equation of the KF algorithm is expressed as:

[0060] (5),

[0061] In formula (5), represents time the error covariance at the epoch moment, represents the corresponding covariance matrix of represents the epoch to the epoch the prior state vector estimate value; represents from the epoch to the epoch the state transition matrix; represents the epoch the state vector estimate value; represents the epoch to the epoch the one-step covariance of the predicted estimation error; represents the epoch the predicted estimation error covariance; represents from the epoch to the epoch the transpose of the state transition matrix; represents the transpose of the noise distribution matrix;

[0062] where the observation update equation of the KF algorithm is expressed as:

[0063] (6),

[0064] In formula (6), represents the gain matrix; represents the transpose of the observation transfer matrix; represents the observation noise covariance; represents the epoch the state vector estimate value; represents the epoch the predicted estimation error covariance;

[0065] Through the observation update equation of the KF algorithm, the position coordinates of the carrier can be updated, and the updated position coordinates of the carrier are fed back to the GNSS. The GNSS corrects the preliminary position coordinates of the carrier in step 3 through the updated position coordinates of the carrier, and finally the corrected position coordinates of the carrier are output as the positioning result.

[0066] On the basis of the above embodiments, in a preferred embodiment, in step 4, after obtaining the observation update equation of the KF algorithm, it further includes the observation noise covariance Adaptive adjustment; the masked image with specific semantic tags obtained in step 1 contains the semantic tag of the occluder. The depth information of the occluder is obtained through the local mapping module of the SLAM system. The depth information of the occluder is the distance between the carrier and the occluder, and the occlusion factor is constructed is:

[0067] (7),

[0068] In formula (7), is the depth of field, and the depth of field ranges from ;

[0069] When an occluder is detected in the image, the occlusion factor is integrated onto the original observation noise covariance to obtain the observation noise covariance after adaptive adjustment , and the adaptive adjustment formula for the observation noise covariance is:

[0070] (8),

[0071] When an occluder is detected, substituting the observation noise covariance after adaptive adjustment into the observation update equation of the KF algorithm can calculate the optimal estimated value of the carrier's pose.

[0072] The size of the gain matrix is affected by equations (3) to (6). Specifically, when the set observation noise variance exceeds the actual noise distribution level, the corresponding value will decrease, resulting in an inappropriate reduction in the uncertainty range of the true value of the state estimate, and further causing bias in the estimation result. On the contrary, if the set is too small, the value will increase, possibly causing the phenomenon of filter divergence. If there are abnormal occluders in the observation data, and the system still uses the original observation noise covariance without timely adjustment, then the influence of the occluder on the filter will not be suppressed, resulting in an unclear filter convergence effect or even filter divergence.

[0073] Therefore, the adaptive adjustment of the original observation noise covariance is to improve the estimation characteristics and robustness of the KF algorithm. Integrating the occlusion factor onto the original observation noise covariance and adjusting upward to obtain the observation noise covariance after adaptive adjustment , thus reducing the gain matrix , reducing the influence of abnormal measurement data on the KF algorithm.

[0074] Specifically, in step 3, the basic observations of GNSS mainly include pseudorange observations and carrier phase observations. According to the definitions of pseudorange observations and carrier phase observations, their positioning equations can be expressed as:

[0075] (9),

[0076] In formula (9), and respectively represent the pseudorange observation value and the carrier phase observation value, is the true satellite-ground distance, is the speed of light in vacuum, and are the GNSS receiver clock error and the satellite clock error respectively, and are the ionospheric refraction error and the tropospheric refraction error respectively, is the carrier wavelength, is the integer ambiguity, and respectively represent the unmodeled residual errors of the pseudorange observation value and the carrier phase observation value.

[0077] The accuracy of pseudorange observations is relatively low, and real-time solution of the integer ambiguity is required for high-precision dynamic positioning using carrier phase observations . For short-baseline positioning, the atmospheric errors between stations are highly correlated. Therefore, the synchronous observation difference method between stations is used to eliminate or reduce the influence of these errors.

[0078] For the case where two receivers simultaneously observe multiple satellites, the calculation formula of the double-difference observation model is expressed as:

[0079] (10),

[0080] where the superscripts and represent the satellite numbers; the subscripts and represent the station numbers.

[0081] The double-difference observation model eliminates the receiver clock error by calculating the difference between the observation values of two receivers for different satellites. In carrier phase double-difference positioning, the elimination of these errors also facilitates the dynamic solution of the integer ambiguity.

[0082] In the method of the present invention, the GNSS receiver independently obtains the position coordinates of the carrier through its double-difference measurement model. At the same time, the continuous image frames captured by the stereo camera are transmitted to the VO module for performing the pose estimation of the carrier. Step 4 of the present invention adopts a loose-coupling model to combine the carrier position coordinates of GNSS and the pose information of the carrier obtained by the VO module, and constructs a loose-coupling positioning system of GNSS / vision. During this period, the GNSS / vision loose-coupling positioning system calculates the occlusion factor according to the environmental semantic information and the local mapping module .

[0083] Subsequently, the GNSS / vision loose-coupling positioning system introduces an adaptively adjusted KF algorithm in combination with the occlusion factor. In this algorithm, the pose output by the VO module is used as the state vector for updating, and the difference between GNSS positioning and visual positioning is used as the observation vector for updating. Under the constraint of the occlusion factor , the adaptive KF algorithm can calculate the optimal estimated value, which is then directly fed back to the GNSS module to correct its position coordinates. Finally, the corrected position coordinates are used as the output result of the integrated system.

[0084] Based on the above embodiments, in a preferred embodiment, the image is preprocessed before being input into the VO module. The preprocessing includes performing distortion correction on the acquired image to correct the distortion effect of the lens distortion on the image, and then removing the random noise in the image through a noise reduction algorithm (such as Gaussian filtering) to improve the clarity and usability of the image.

[0085] Based on the above embodiments, in a preferred embodiment, in step 1, the environmental semantic segmentation processing unit uses the YOLOv5 method to perform environmental semantic segmentation on the image.

[0086] The above are only the specific implementation manners of the present invention, enabling those skilled in the art to understand or implement the present invention. Although the above embodiments have been described in detail, those of ordinary skill in the art should understand that they can still modify the technical solutions described in the above embodiments, or perform equivalent replacements on some or all of the technical features; and these modifications or replacements do not cause the essence of the corresponding technical solutions to deviate from the scope of the technical solutions of the above embodiments, and they should all be covered by the protection scope of the claims.

Claims

1. A GNSS / vision loosely coupled positioning method integrating environmental semantics, characterized in that The loose-coupling positioning method is implemented based on a stereo camera, GNSS, and SLAM system. The stereo camera is fixedly connected to the carrier to collect images during the carrier's driving in real time. The SLAM system includes a VO module and a local mapping module. The VO module includes a feature point tracking unit. The steps of the loose-coupling positioning method are as follows: Step 1: Improve the VO module by embedding a real-time environmental semantic segmentation processing unit into the VO module. The environmental semantic segmentation processing unit works in parallel with the feature point tracking unit. Images are input into the VO module in the form of continuous frames. The feature point tracking unit extracts ORB feature points from the images to form a feature map containing dynamic and static feature points. The environmental semantic segmentation processing unit performs environmental semantic segmentation on the images to generate a masked image with specific semantic labels. The VO module fuses the information of the feature map and the masked image, marks the dynamic feature points as outliers, and removes the outliers to obtain an image with only static feature points. Step 2: The VO module performs point-by-point matching between the static feature points of the current image and the static feature points of the previous frame, and then obtains the six-degree-of-freedom rotation R and translation T of the stereo camera by minimizing the reprojection error. Step 3: Independently obtain the preliminary position coordinates of the carrier based on the GNSS double-difference observation model. Step 4: The rotation R and translation T of the stereo camera are the rotation R and translation T of the carrier. Use the rotation R and translation T of the carrier to construct a visual positioning error model, where the visual positioning error model is: （1）， In Equation (1), , and are the length scale factor of the stereo camera, the position coordinates of the vehicle, and the attitude of the vehicle in the navigation coordinate system, respectively. The position coordinates are obtained by integrating the translation T of the vehicle, and the attitude is obtained by integrating the rotation R of the vehicle; , and are the errors of the length scale factor, position coordinates, and attitude in the navigation coordinate system, respectively; , and are the continuous-time derivatives of the errors of the length scale factor, position coordinates, and attitude in the navigation coordinate system, respectively; is the estimated rotation matrix from the vehicle coordinate system to the navigation coordinate system; is the estimated translation in the vehicle coordinate system; is the estimated value of the length scale factor; and are the translation error and rotation error in the vehicle coordinate system, respectively; is 's random drift; Simplify formula (1) into a system model, and the system model is expressed as: （2）， In formula (2), is ; is ; is ; is ; is ; Then construct the state model form of the KF algorithm through formula (2): （3）， In formula (3), is the state vector at epoch , is the discrete-time transition matrix from epoch to epoch , is the noise distribution matrix, is the system noise sequence; Adopt a loose-coupling model, and use the difference between the position coordinates of the carrier obtained by the VO module and the preliminary position coordinates of the carrier obtained by GNSS in step 3 as the observation value input of the KF algorithm to construct the observation model of the KF algorithm as: （4）， In formula (4), and are the carrier position coordinates obtained by the VO module and GNSS respectively; is the observation transfer matrix; is the observation noise; The state update equation of the KF algorithm is expressed as: (5), In formula (5), represents time the error covariance at the epoch time, represents the corresponding covariance matrix of; represents the epoch to the epoch the prior state vector estimate; represents from the epoch to the epoch the state transition matrix; represents the epoch the state vector estimate; represents the epoch to the epoch the one-step covariance of the prediction estimation error; represents the epoch the prediction estimation error covariance; represents from the epoch to the epoch the transpose of the state transition matrix; represents the transpose of the noise distribution matrix; Among them, the observation update equation of the KF algorithm is expressed as: （6）， In formula (6), represents the gain matrix; represents the transpose of the observation transition matrix; represents the observation noise covariance; represents the epoch estimated value of the state vector; represents the epoch predicted estimated error covariance; The position coordinates of the carrier can be updated through the observation update equation of the KF algorithm, and the updated position coordinates of the carrier are fed back to GNSS. GNSS corrects the preliminary position coordinates of the carrier in step 3 through the updated position coordinates of the carrier, and finally the corrected position coordinates of the carrier are output as the positioning result.

2. The GNSS / vision loosely coupled positioning method integrating environmental semantics according to claim 1, characterized in that, In step 4, after obtaining the observation update equation of the KF algorithm, it further includes the adaptive adjustment of the observation noise covariance ; the masked image with specific semantic labels obtained in step 1 contains the semantic label of the occluder. The depth information of the occluder is obtained through the local mapping module of the SLAM system. The depth information of the occluder is the distance between the carrier and the occluder, and an occlusion factor is constructed as follows: （7）， In formula (7), is the depth of field, and the depth of field ranges from ; When it is detected that the image contains an occluder, the occlusion factor is integrated into the original observation noise covariance to obtain the observation noise covariance after adaptive adjustment . The adaptive adjustment formula for the observation noise covariance is as follows: （8）， When an occluder is detected, the adaptively adjusted observation noise covariance is brought into the observation update equation of the KF algorithm, and the optimal estimated value of the pose of the vehicle can be calculated.

3. The GNSS / vision loosely coupled positioning method integrating environmental semantics according to claim 1, characterized in that The images are preprocessed before being input into the VO module. The preprocessing includes performing distortion removal on the acquired images, and then removing random noise in the images through a noise reduction algorithm.

4. The GNSS / vision loosely coupled positioning method integrating environmental semantics according to any one of claims 1 to 3, characterized in that, In step 1, the environmental semantic segmentation processing unit uses the YOLOv5 method to perform environmental semantic segmentation on the images.

Images