Vision inertial navigation auxiliary method and device for detecting and repairing compass satellite cycle slip

Abstract

The application provides a Beidou satellite cycle slip detection and repair method and device assisted by visual inertial navigation, wherein the method comprises the following steps: predicting a carrier prior position by using acceleration and angular velocity provided by an inertial measurement unit through dead reckoning; acquiring an original image, performing feature extraction and matching on the original image, and dividing visual features of the original image into long-term tracking features and short-term tracking features; combining a mixed updating mode of long-term and short-term visual features, constructing an observation equation of visual feature points, and optimizing inertial navigation prior position accuracy and visual feature state quantity; and constructing a cycle slip detection quantity with a longer wavelength by using linear combination and non-combination phase observation values, judging whether a cycle slip occurs or not, and performing cycle slip repair. Through the application, high-precision and high-reliable positioning of a low-cost multi-source fusion system in a complex urban scene can be realized, and the problem of poor accuracy and reliability existing in the prior art is solved.

Description

Technical Field

[0001] This invention relates to the field of navigation and positioning technology, and in particular to a method and apparatus for detecting and repairing cycle slips of BeiDou satellites with visual inertial navigation assistance. Background Technology

[0002] Cycle slip detection and repair, which can screen good carrier phase observations for ambiguity resolution and fixation, is one of the key technologies for achieving high-precision positioning with BeiDou satellites. Early studies used only BeiDou satellite observations for cycle slip detection and repair. However, observations without ionosphere amplify residual phase errors, reducing the accuracy of cycle slip detection. Pseudorange phase combination observations introduce significant pseudorange noise and multipath propagation, making it difficult to detect small cycle slips. Polynomial fitting and high-order differences between epochs are easily affected by high-frequency noise, limiting cycle slip detection performance. Doppler methods and three-frequency observations place high demands on receivers, making them difficult to adapt to low-cost receivers with poor data quality. These methods perform well under static conditions with good observation environments; however, in dynamic and complex urban scenarios, their performance rapidly declines due to severe satellite signal obstruction and frequent loss of lock, with a more significant impact on low-cost receivers and antennas with high observation noise and poor signal quality.

[0003] Inertial sensors (INS) possess the advantages of being autonomous, passive, and unaffected by external environments. They provide accurate relative positions in the short term to assist in cycle slip detection and repair, and their recursive absolute position accuracy is generally superior to pseudorange. Existing research has used the relative positions provided by INS to construct differential phase observations between phase epochs for cycle slip detection, verifying that this method can effectively detect single-frequency cycle slips. However, this method relies on good phase observations from the previous epoch. Some researchers have used the absolute positions provided by INS to construct cycle slip probes, and combined with satellite elevation angles and geometry, can achieve accurate detection of single-frequency phase cycle slips in the current epoch. To address the problem of rapid error accumulation in low-cost INS, some studies have used dual-frequency linear combination observations to construct cycle slip probes, which can construct longer wavelengths and improve tolerance to INS recursive position errors. However, linear combination observations amplify residual phase errors, and when calculating single-frequency phase cycle slips based on cycle slips from combined observations, these residual errors are further amplified, leading to a decrease in the reliability of cycle slip repair.

[0004] Visual sensors can perceive rich texture information in the surrounding environment to calculate the relative pose of the carrier. Combined with inertial navigation, visual-inertial odometry (VIO) enables low-drift dead reckoning to provide more accurate prior positions. Its approach is essentially the same as inertial navigation-assisted cycle slip detection, utilizing visual-inertial odometry to provide higher-precision relative or absolute positions to construct probes for cycle slip detection and correction. Existing research has verified that visual-inertial navigation can be used to provide more accurate relative positions for single-station, single-frequency cycle slip detection. Some researchers have also studied visual-inertial navigation prior position information-assisted precise point positioning (PPP) methods for cycle slip detection, ambiguity resetting, and multipath estimation. Related studies typically employ Multi-State Constraint Kalman Filter (MSCKF) to fuse visual observation information. This method breaks down long-term tracked visual features into multiple short-term tracked visual features for processing. It fails to fully exploit the co-view relationship across multiple frames to constrain the accumulation of inertial navigation errors, making it difficult to suppress the rapid accumulation of low-cost inertial navigation errors. Furthermore, the provided prior position accuracy is not high, resulting in limited improvement in cycle slip detection and repair. In addition, the use of phase observations is primarily single-frequency, without utilizing dual-frequency observations for linear combination. This leads to a decline in cycle slip detection performance under long-term BeiDou satellite signal loss.

[0005] In summary, existing visual-inertial navigation (VIS)-assisted cycle slip detection methods for the BeiDou satellite system are quite similar to those for inertial navigation-assisted methods. Visual observation information is not fully utilized to constrain the rapid accumulation of errors in low-cost inertial navigation systems, making it difficult to accurately detect small cycle slips even during long-term signal loss. Furthermore, existing studies often use linear combinations of observations for cycle slip detection, but this method amplifies phase noise, leading to a decrease in the reliability of cycle slip repair. Therefore, this invention proposes a visual-inertial navigation (VIS)-assisted BeiDou satellite cycle slip detection and repair method. It employs a hybrid update method of long-term and short-term visual features to improve the accuracy maintenance capability of the visual-inertial odometry when the satellite signal is lost. Simultaneously, it adopts an adaptive switching strategy between linear and non-combined phase observations, enabling simultaneous handling of single-frequency and multi-frequency cycle slips, achieving accurate detection and reliable repair of frequent cycle slips in complex urban scenarios.

[0006] There is currently no effective solution to the problems of poor accuracy and reliability of existing related technologies. Summary of the Invention

[0007] This invention provides a method and apparatus for detecting and repairing BeiDou satellite cycle slips with visual inertial navigation assistance, in order to solve the defects of poor accuracy and reliability in existing related technologies.

[0008] In a first aspect, the present invention provides a method for visual-inertial navigation-assisted BeiDou satellite cycle slip detection and repair, comprising: The acceleration and angular velocity provided by the inertial measurement unit are obtained, and the prior position of the vehicle is predicted by dead reckoning. The original image is acquired, and features are extracted and matched on the original image. The visual features of the original image are divided into long-term tracking features and short-term tracking features. By combining long-term and short-term visual features in a hybrid update method, an observation equation for visual feature points is constructed to optimize the accuracy of inertial navigation prior pose and visual feature state variables. A longer-wavelength cycle slip detection is constructed using linear and non-linear phase observations to determine whether a cycle slip has occurred and to perform cycle slip repair.

[0009] According to the present invention, a method for detecting and repairing BeiDou satellite cycle slips with visual inertial navigation assistance is provided, which acquires acceleration and angular velocity provided by an inertial measurement unit and predicts the prior position of the carrier through dead reckoning, including: The acceleration and angular velocity provided by the inertial measurement unit are obtained, and the prior position and its variance at the current moment are determined by dead reckoning.

[0010] According to a visual-inertial navigation-assisted method for BeiDou satellite cycle slip detection and repair provided by the present invention, an original image is acquired, and feature extraction and matching are performed on the original image. The visual features of the original image are divided into long-term tracking features and short-term tracking features, including: The original image is acquired, features are extracted and matched on the original image, and poor-quality outliers are removed using the random sampling consistency method. The visual feature points of each frame of the image are obtained and the visual features are divided into long-term tracking features and short-term tracking features based on the number of consecutive tracking frames.

[0011] According to the present invention, a method for cycle slip detection and repair of BeiDou satellites with visual inertial navigation assistance is provided. This method combines a hybrid update method of long-term and short-term visual features to construct observation equations for visual feature points, optimizing the prior pose accuracy of the inertial navigation system and the visual feature state variables. The method includes: For long-term tracking features, determine whether the long-term tracking feature is a new visual feature point or an old visual feature point, and update it according to the determination result; For short-term tracking features, after constructing the visual observation equation, the short-term tracking features are eliminated from the parameters to be estimated by using left null space projection. During measurement updates, only the inertial navigation pose is updated, and the visual features are not updated.

[0012] According to a visual-inertial navigation-assisted method for detecting and repairing cycle slips in BeiDou satellites provided by the present invention, the method determines whether the long-term tracking feature is a new visual feature point or an old visual feature point, and updates it based on the determination result, including: If the feature points of the long-term tracking feature are not added to the state vector, then the long-term tracking feature is a new visual feature point. The new visual feature point is used as the state to be optimized through state augmentation, and the covariance matrix is ​​calculated. The new visual feature point is then added to the state vector. If the feature points of the long-term tracking feature have been added to the state vector, then the long-term tracking feature is the old visual feature point. The new visual observation value of the old visual feature point is used to construct the observation equation about the inertial navigation pose and visual feature points within the window for extended Kalman filter update, and the inertial navigation pose and visual feature points are updated at the same time.

[0013] According to the present invention, a method for detecting and repairing cycle slips of BeiDou satellites with visual-inertial navigation assistance is provided. This method constructs cycle slip detection measurements using linearly combined and non-combined phase observations to determine whether a cycle slip has occurred, including: The phase observations provided by the base station and the rover are obtained to construct the cycle slip detection measurement. Based on the prior position variance and phase noise, the cycle slip detection threshold is determined. Compare the cycle slip measurement with the cycle slip detection threshold. If the cycle slip measurement exceeds the cycle slip detection threshold, it is determined that a cycle slip has occurred.

[0014] If the satellite signal loss time is long, use phase observations at different frequencies to make a linear combination to construct a linear combination cycle slip measurement. Compare the linear combination cycle slip measurement with the cycle slip detection threshold. If the linear combination cycle slip measurement is less than the cycle slip detection threshold, it is considered that no cycle slip has occurred, and a double-difference observation equation is constructed for measurement update. If the linear combination cycle slip detection measurement reaches the cycle slip detection threshold, then a cycle slip is considered to have occurred.

[0015] According to the present invention, a method for detecting and repairing cycle slips of BeiDou satellites with visual inertial navigation assistance is provided. If a cycle slip occurs and the decimal part of the cycle slip detection exceeds a preset number of cycles, the deviation is considered too large to be repaired, and the ambiguity of all frequency points of the satellite is reset. If a cycle slip occurs, and the decimal part of the cycle slip probe does not exceed the preset number of cycles, then a linear combination and non-combination mode is used for cycle slip repair.

[0016] According to the present invention, a method for detecting and repairing cycle slips of BeiDou satellites with visual inertial navigation assistance is provided. If the cycle slip of a single frequency point calculated by linear combination is unreliable, the cycle slip value of the single frequency point is calculated using a non-combination mode. The cycle slip value is then corrected to the ambiguity parameter and the state variance is expanded. Finally, a double-difference observation equation is constructed for quality control and measurement updates.

[0017] Secondly, the present invention also provides a visual-inertial navigation-assisted device for detecting and repairing BeiDou satellite cycle slips, comprising: The preprocessing module is used to obtain the acceleration and angular velocity provided by the inertial measurement unit, construct cycle slip detection measurements in combination with the prior position, and determine whether a cycle slip has occurred. The extraction module is used to acquire the original image, extract and match features from the original image, and divide the visual features of the original image into long-term tracking features and short-term tracking features. The update module is used to construct the observation equation of visual feature points by combining a hybrid update method of long-term and short-term visual features, and to optimize the inertial navigation pose and visual feature state variables. The repair module is used to construct a longer-wavelength cycle slip probe using linear and non-linear phase observations, determine whether a cycle slip has occurred, and perform cycle slip repair.

[0018] Thirdly, the present invention also provides an electronic device, including a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the program to implement the visual-inertial navigation-assisted BeiDou satellite cycle slip detection and repair method as described in the first aspect above.

[0019] Fourthly, the present invention also provides a non-transitory computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the method for visual-inertial navigation-assisted BeiDou satellite cycle slip detection and repair as described in the first aspect above.

[0020] Fifthly, the present invention also provides a computer program product, including a computer program that, when executed by a processor, implements the visual-inertial navigation-assisted BeiDou satellite cycle slip detection and repair method as described in the first aspect above.

[0021] Compared with the prior art, the present invention has the following beneficial effects: This invention provides a visual-inertial navigation-assisted method for BeiDou satellite cycle slip detection and repair. By constructing a longer wavelength cycle slip probe, the impact of recursive position errors on cycle slip detection is reduced. Simultaneously, the low noise of non-combined observations enables accurate repair of cycle slips at a single frequency. This method fully utilizes the observation information from low-cost sensors, significantly improving the accuracy and reliability of cycle slip detection and repair when BeiDou signal quality deteriorates. Furthermore, it can simultaneously handle single-frequency and multi-frequency phase cycle slips of BeiDou satellites, thereby achieving high-precision and high-reliability positioning for low-cost multi-source fusion systems in complex urban scenarios, solving the problems of poor accuracy and reliability in existing related technologies. In addition, this method can also be extended to global navigation satellite systems. Attached Figure Description

[0022] To more clearly illustrate the technical solutions in this invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are some embodiments of this invention. For those skilled in the art, other drawings can be obtained from these drawings without creative effort.

[0023] Figure 1 This is a flowchart of the method for visual inertial navigation-assisted BeiDou satellite cycle slip detection and repair provided by the present invention; Figure 2 This is a schematic diagram of the visual-inertial navigation-assisted BeiDou satellite cycle slip detection process in an embodiment of the present invention; Figure 3 This is a schematic diagram of the long-term and short-term visual feature classification strategy in an embodiment of the present invention; Figure 4 This is a structural block diagram of the device for visual inertial navigation-assisted BeiDou satellite cycle slip detection and repair provided by the present invention; Figure 5 This is a schematic diagram of the structure of the electronic device provided by the present invention. Detailed Implementation

[0024] To make the objectives, technical solutions, and advantages of this invention clearer, the technical solutions of this invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some, not all, of the embodiments of this invention. All other embodiments obtained by those skilled in the art based on the embodiments of this invention without creative effort are within the scope of protection of this invention.

[0025] This invention provides a method for visual-inertial navigation-assisted cycle slip detection and repair of BeiDou satellites. Figure 1 This is a flowchart of the visual-inertial navigation-assisted BeiDou satellite cycle slip detection and repair method provided by the present invention, as shown below. Figure 1As shown, the method includes the following steps: Step S101: Obtain the acceleration and angular velocity provided by the Inertial Measurement Unit (IMU), and predict the prior position of the vehicle through dead reckoning; Step S102: Obtain the original image, perform feature extraction and matching on the original image, and divide the visual features of the original image into long-term tracking features and short-term tracking features; Step S103: Combining the hybrid update method of long-term and short-term visual features, construct the observation equation of visual feature points and optimize the inertial navigation prior pose accuracy and visual feature state quantity. Step S104: Construct a longer-wavelength cycle slip probe using linear and non-linear phase observations to determine whether a cycle slip has occurred and perform cycle slip repair.

[0026] In this method, firstly, the acceleration and angular velocity provided by the inertial measurement unit (IMU) are acquired, and the prior position of the carrier is predicted through dead reckoning. Then, the original images are acquired, and feature extraction and matching are performed on them, dividing the visual features of the original images into long-term tracking features and short-term tracking features. Based on this, a hybrid update method combining long-term and short-term visual features is used to construct the observation equation for visual feature points, optimizing the accuracy of the IMU's prior pose and the visual feature state variables. The multi-frame co-view relationship of long-term tracking features is fully utilized to constrain the rapid accumulation of errors in low-cost IMUs, improving the accuracy of the prior position. Finally, cycle slip detection is constructed using linearly combined and non-combined phase observations to determine whether a cycle slip has occurred and to perform cycle slip repair. By constructing a cycle slip detection with a longer wavelength, the influence of recursive position errors on cycle slip detection is reduced. Simultaneously, the advantage of low noise from non-combined observations is utilized to achieve accurate repair of cycle slips at a single frequency point. Through the above process, the observation information of low-cost sensors can be fully utilized, significantly improving the accuracy and reliability of cycle slip detection and repair when BeiDou signal quality deteriorates. Furthermore, it can simultaneously handle single-frequency and multi-frequency phase cycle slips of BeiDou satellites, thereby achieving high-precision and high-reliability positioning for low-cost multi-source fusion systems in complex urban scenarios, solving the problems of poor accuracy and reliability in existing related technologies. In addition, this method can also be extended to global navigation satellite systems.

[0027] In some of these embodiments, such as Figure 2 As shown, Figure 2 This is a schematic diagram of the BeiDou satellite cycle slip detection process assisted by visual inertial navigation in an embodiment of the present invention. Step S101, obtaining the acceleration and angular velocity provided by the inertial measurement unit, and predicting the prior position of the carrier through dead reckoning, includes: obtaining the acceleration and angular velocity provided by the inertial measurement unit, and determining the prior position and its variance at the current moment through dead reckoning.

[0028] In this embodiment, based on the prior position provided by visual inertial navigation, the double-difference residual of the carrier phase observation at a certain frequency is calculated to construct a cycle slip detection measurement. The cycle slip detection threshold is then calculated in real time based on the prior position variance and phase noise. If the cycle slip detection measurement exceeds the threshold, the phase observation is considered to have experienced a cycle slip. The calculation of the double-difference residual of the carrier phase observation using the prior position provided by visual inertial navigation is more accurate than the position calculated by pure inertial navigation. The real-time calculation of the cycle slip detection threshold based on the prior position variance and phase noise allows for adaptation to different scenarios and lockout durations compared to fixed empirical thresholds.

[0029] In some of these embodiments, Figure 3 This is a schematic diagram of the long-term and short-term visual feature classification strategy in an embodiment of the present invention, such as... Figure 3 As shown, step S102 involves acquiring the original image, performing feature extraction and matching on the original image, and dividing the visual features of the original image into long-term tracking features and short-term tracking features. This includes: acquiring the original image, performing feature extraction and matching on the original image, and using a random sampling consistency method to remove visual outliers with poor quality; acquiring the observation values ​​of visual feature points of each frame image, and dividing the visual features into long-term tracking features and short-term tracking features based on the number of consecutive tracking frames.

[0030] Based on this, in step S103, the observation equation of the visual feature point is constructed by combining the hybrid update method of long-term and short-term visual features, and the accuracy of the inertial navigation prior pose and the visual feature state variables are optimized. This includes: for long-term tracking features, determining whether the long-term tracking features are new visual feature points or old visual feature points, and updating them according to the determination result; for short-term tracking features, after constructing the visual observation equation, the short-term tracking features are eliminated from the parameters to be estimated by using the left null space projection method. During measurement update, only the inertial navigation pose is updated, and the visual features are not updated, so as to avoid the increase in the dimension of the state variables and the amount of computation.

[0031] Specifically, the process involves determining whether a long-term tracking feature is a new visual feature point or an old visual feature point, and updating it accordingly. This includes: if the feature point of the long-term tracking feature has not been added to the state vector, then the long-term tracking feature is a new visual feature point. The new visual feature point is then used as the state to be optimized through state augmentation, and the covariance matrix is ​​calculated before adding the new visual feature point to the state vector. If the feature point of the long-term tracking feature has already been added to the state vector, then the long-term tracking feature is an old visual feature point. The new visual observations of the old visual feature points are used to construct the observation equations about the inertial navigation pose and visual feature points within the window, and then the Extended Kalman Filter (EKF) is used for updating. At the same time, the inertial navigation pose and visual feature points are updated.

[0032] In this embodiment, visual features are divided into long-term and short-term tracking features based on the number of tracking frames. Long-term tracking features are added to the state variables as parameters to be optimized through visual feature state augmentation. The multi-frame co-view relationship is fully utilized to constrain the rapid accumulation of errors in low-cost inertial navigation and improve the ability to maintain position accuracy when BeiDou satellite signals are lost.

[0033] First, a long-term and short-term visual feature classification and management strategy is implemented. Based on the number of tracking frames, visual features are divided into long-term and short-term tracking features, thus achieving a classification and management strategy based on long-term and short-term visual features. Features with more tracking frames than the number of cameras within the window are considered long-term tracking features. These are used as parameters to be optimized through visual feature state augmentation, which can fully utilize multi-frame co-view relationships to improve positioning accuracy. Features with fewer tracking frames than the number of cameras within the sliding window are considered short-term tracking features. These are discarded after constructing the visual observation equation, avoiding a rapid increase in the dimensionality of the state variables and ensuring computational efficiency. Therefore, the system state variables can be represented as:

[0034] in, x This represents system state variables, including inertial navigation state, inertial navigation pose and visual feature point state within the sliding window, and carrier phase double-difference ambiguity parameters. This represents the 15-dimensional inertial navigation state, corresponding sequentially to the position. ,speed Posture Quaternions Additional calculation Zero bias of the gyroscope ; This represents the inertial navigation pose corresponding to each camera frame within the sliding window. Indicates the first i The inertial navigation pose corresponding to each frame camera moment. i Represents a camera frame. N Indicates the number of camera frames within the window; This represents the long-term tracking features within the sliding window. M Indicates the number of visual features within the window. Indicates the first j The coordinates of the location of each visual feature point; This represents the carrier phase double-difference ambiguity parameter. Indicates the first k A fuzziness parameter, n This represents the number of ambiguity parameters.

[0035] Secondly, there is visual feature state augmentation. Long-term tracked features are added to the state variables as parameters to be optimized through state augmentation to fully utilize their co-viewing relationships across multiple frames. The covariance matrix of the newly added visual feature states can be written in the following form:

[0036] in, Describe the covariance matrix. The state covariance matrix before augmentation includes 15-dimensional inertial navigation pose, inertial navigation pose at historical frame times, and existing visual features and carrier phase double-difference ambiguity parameters within the window. and These represent the covariance of the newly added visual features and the covariance terms with other state variables, respectively.

[0037] Finally, a hybrid visual update method is used. Observation equations are constructed using visual observations from both long-term and short-term features. Measurement updates and closed-loop feedback of state variables constrain the rapid divergence of inertial navigation (INS) errors. Short-term tracking features construct visual observation equations for the INS poses of all frames within the window; while long-term tracking features utilize observation information from all historical frames and can construct visual observation equations for the INS poses and visual feature points within the window using the observations of the latest frame. This fully leverages multi-frame co-view relationships to reduce triangulation errors at these feature points. Based on the pinhole camera projection model, the visual observation equations can be written in the following form:

[0038] in, Represents pixel coordinates, Indicates that feature points were observed. A certain camera frame, Indicates visual feature points in the observation camera system ( The depth of the system Representing feature points In the observation camera system ( 3D coordinates under (system) , and They represent and Inertial navigation pose at frame time. K Indicates camera intrinsic parameters. This indicates the camera's extrinsic parameters, which can be obtained through offline calibration. Indicates the feature point in the geocentric geofixed coordinate system ( e 3D coordinates under (system) By calculating the Jacobian matrix of the inertial navigation pose state and visual feature state corresponding to the camera frame time using the perturbation method, the linearized visual observation equation is obtained:

[0039] in, and Let represent the Jacobian matrices of the inertial navigation pose and the visual feature points within the window, respectively. Indicates the first [number]th ... iFrame inertial navigation attitude quaternion error Indicates the first [number]th ... a The inertial navigation attitude quaternion error for a frame (usually the first frame) is calculated. The above visual observation equations establish the constraint relationship between visual observations and all visual state quantities, which can make full use of visual co-viewing relationships to effectively constrain the accumulation of inertial navigation errors.

[0040] For example, if the maximum number of camera frames within the sliding window is N, the number of tracking frames for a certain visual feature point includes the following four cases: Case 1: If the number of tracking frames for a feature point is less than N, it indicates that the feature point was lost during tracking in the current frame and is not a long-term tracked visual feature. This feature point is used to construct the observation equation for the inertial navigation pose within the window and to perform Multi-State Constraint Kalman Filter (MSCKF) updates. Case 2: The number of tracking frames for a feature point exceeds N, and the feature point has not yet been added to the state vector; this is a new visual feature point that has been tracked for a long time. State augmentation is used to treat this feature point as a state to be optimized, and its covariance matrix is ​​calculated and added to the state vector. Case 3: The number of tracking frames for a feature point exceeds N, and the feature point has already been added to the state vector, i.e., it is an old visual feature point that has been tracked for a long time. Using the new visual observations of this feature point, an observation equation for the inertial pose and visual feature point within the window is constructed for EKF update, simultaneously updating the inertial pose and visual feature point; Case 4: If a visual feature point that already exists in the state variable is lost in the current frame, it is removed from the state variable by marginalization and its corresponding covariance matrix is ​​cleared to avoid the state variable dimension from increasing continuously. Furthermore, if the number of visual feature points tracked over a long period exceeds the maximum number of feature points M within the window, the excess points are still handled according to Case 1 to avoid a rapid increase in the dimension of the state variables and a reduction in computational efficiency.

[0041] After classifying short-term and long-term visual feature points, visual features for short-term and long-term tracking are obtained. A hybrid visual update method is adopted to fully utilize the multi-frame co-view relationship of long-term tracking visual features while ensuring computational efficiency, thereby improving the accuracy maintenance capability of visual inertial odometry. Specifically, the visual features for short-term and long-term tracking are processed as follows.

[0042] Case 1: For short-term tracking feature points The 3D coordinates of the feature points in the reference camera frame are obtained through feature point triangulation. and construct its information about the first i The visual observation equation for the inertial navigation pose corresponding to a frame of the camera is as follows:

[0043] in, Represents pixel coordinates, Representing feature points In the i Depth in the camera coordinate system and They represent the first i Frame camera and reference camera frame. The above observation equations are linearized using the perturbation method, and can be written in the following form:

[0044] in, and This represents the observed and calculated pixel coordinates of the feature point. This indicates the reprojection error. This is visual observation noise; and This represents the inertial navigation pose and visual feature point state quantities corresponding to each camera frame. and These are the corresponding Jacobian matrices, in the following forms:

[0045] in:

[0046] in, and This refers to the camera's intrinsic parameters, which can be obtained through offline calibration. Indicates the first in the window i The rotation matrix of the frame inertial navigation system. Indicates visual feature points at e The coordinates of the position are below. Indicates the first in the window i Position of frame inertial navigation Indicates the first in the window a The rotation matrix of the inertial navigation pose for a frame (usually the first frame). Indicates the visual feature point at the th a The position coordinates of the camera in the frame system. For all camera frames that observed this feature point, construct the observation equations, and then solve all the observation equations simultaneously to obtain:

[0047] in, This represents the inertial navigation pose state corresponding to all camera frames that observed this feature point. This represents the observation noise of the visual feature points. The parameters to be estimated in the above observation equation also include the state of the visual feature points. This would lead to a rapid increase in the dimension of the state variables, so it is necessary to remove them from the observation equations by using left null space projection:

[0048] in, Jacobian matrix The left null space is obtained through matrix QR decomposition. This represents the reprojection error after null-space projection. Let represent the Jacobian matrix after null projection. Solve the observation equations corresponding to all short-term tracking visual features simultaneously:

[0049] in, This represents the reprojection error of all short-term tracked visual features. n o Represents the observation noise for all short-term tracking visual features. x clone This represents the inertial navigation pose corresponding to the camera frame time within the sliding window. The above observation equation is applied to MSCKF measurements, optimizing only the inertial navigation pose within the window.

[0050] Scenario 2: For long-term tracking feature points The state vector is augmented with feature point states and its corresponding covariance matrix is ​​calculated. The visual observation equation is constructed and then transformed into the following form using QR decomposition:

[0051] Among them, subscript init The subscript represents the observation equation used for augmenting the state of feature points. upt This represents the observation equation used for measurement updates. Typically, it is an invertible upper triangular matrix. The first row of observation equations is extracted:

[0052] The state-enhanced covariance matrix can be written as:

[0053] in, The covariance matrix is ​​represented by the following specific elements:

[0054] in, This represents the visual observation noise used for feature point state augmentation after QR decomposition. for The corresponding observation noise covariance matrix. The above formula realizes state augmentation for long-term tracking of visual features. Using the new visual observations of this feature in the next frame, the following visual observation equation is constructed:

[0055] in, This represents the Jacobian matrix containing the inertial navigation pose and visual features within the window. The EKF update is performed using the above formula, simultaneously optimizing the inertial navigation pose and visual features within the window. This fully utilizes the multi-frame co-view relationship of long-term tracked visual features, improving the accuracy maintenance capability of the visual inertial odometry.

[0056] In some embodiments, step S104, which involves constructing cycle slip detection measurements using linearly combined and non-combined phase observations, and determining whether a cycle slip has occurred, includes: acquiring phase observations provided by the base station and rover, constructing cycle slip detection measurements, and determining a cycle slip detection threshold based on prior position variance and phase noise; comparing the cycle slip detection measurements with the cycle slip detection threshold, and determining that a cycle slip has occurred if the cycle slip detection measurements exceed the cycle slip detection threshold; if the satellite signal loss time is long, using phase observations at different frequencies to perform linear combination to construct linear combination cycle slip detection measurements; comparing the linear combination cycle slip detection measurements with the cycle slip detection threshold, and considering that no cycle slip has occurred if the linear combination cycle slip detection measurements are less than the cycle slip detection threshold, and constructing a double-difference observation equation for measurement updates; and considering that a cycle slip has occurred if the linear combination cycle slip detection measurements reach the cycle slip detection threshold.

[0057] Based on this, if a cycle slip occurs and the decimal part of the cycle slip probe exceeds a preset number of cycles, the deviation is considered too large to be repaired, and ambiguity is reset for all frequency points of the satellite; if a cycle slip occurs and the decimal part of the cycle slip probe does not exceed the preset number of cycles, cycle slip repair is performed using linear combination and non-combination modes. Preferably, the preset number of cycles is 0.25 cycles.

[0058] Furthermore, if the cycle slip of a single frequency point calculated based on the linear combination is unreliable, the cycle slip value of a single frequency point is calculated using the non-combination mode, the cycle slip value is corrected to the ambiguity parameter and the state variance is expanded, and then a double-difference observation equation is constructed for quality control and measurement updates.

[0059] In this embodiment, during the cycle slip detection stage, multi-frequency phase observations are linearly combined to construct a cycle slip detection measurement with a longer wavelength, thereby increasing the tolerance to prior position errors. During the cycle slip repair stage, if the combined observations fail to repair the cycle slip, the system switches to non-combined observations to repair the phase cycle slip at a single frequency point, thereby reducing the impact of the large phase noise and multipath of the combined observations on the cycle slip repair performance.

[0060] During cycle slip detection, if prior position errors accumulate significantly, relying solely on single-frequency phase observations is insufficient to determine small cycle slips. By constructing linear combinations of different frequencies, longer-wavelength phase observations can be obtained, thereby reducing the impact of prior position errors on cycle slip detection measurements. Taking dual-frequency observations as an example, the linear combination observations are defined as follows:

[0061] in, and Indicates the frequencies of the first and second frequency points. and The corresponding phase observation value is in meters. and Represents the coefficients of a linear combination; For the wavelength of the combined observations, These are combined observations.

[0062] Based on the phase double-difference observation equation, the following cycle slip detection statistic is constructed using a linear combination of observations:

[0063] in, For the cycle slip measurement corresponding to the combined observation values, and These represent the double-difference phase observations and the double-difference satellite-to-Earth geometric distance, respectively. Indicates double-difference ambiguity. This represents the double-difference phase residual error, including phase noise and multipath, where a and b represent the base station and rover station, respectively, and m and k represent the non-reference satellite and reference satellite, respectively. Once the ambiguity of the previous epoch is successfully fixed, if the cycle slip measurement exceeds the threshold, it is considered that cycle slips may occur at either of these frequency points.

[0064] Since cycle slip detection measurements contain prior position and phase residual errors, relying solely on the measurement magnitude is insufficient to determine whether a cycle slip has occurred. Therefore, an appropriate cycle slip detection threshold needs to be set to assist in the judgment. If no cycle slip has occurred, the cycle slip measurement data follows a zero-mean Gaussian distribution:

[0065] in, The variance corresponding to the phase double difference is set according to the phase noise level; The variance of the double-difference satellite-to-ground distance primarily originates from the prior position variance, provided by the covariance matrix of the inertial navigation state variables. The cycle slip detection threshold can be set as follows:

[0066] in, Indicates the cycle slip detection threshold. This represents the standard deviation of the cycle slip detection measurement. If a cycle slip occurs at the current epoch, the cycle slip detection measurement will show a significant jump. This jump is compared with the cycle slip detection threshold to determine whether a cycle slip has occurred at the current epoch. If the detection measurement exceeds the threshold, a cycle slip is considered to have occurred; otherwise, no cycle slip is considered to have occurred.

[0067] Because linear combination observations amplify residual phase errors, making it difficult to accurately calculate the phase cycle slip value at a single frequency point, non-combined observations are used to correct cycle slips for each frequency point after determining that a cycle slip has occurred in the phase observations. Taking the first frequency point as an example, the formula for calculating cycle slip detection is as follows:

[0068] in, This indicates the frequency jump measurement at the first frequency point. This indicates the carrier phase wavelength at the first frequency point.

[0069] If a cycle slip is determined to have occurred at a given epoch based on cycle slip probe measurements and threshold values, then the cycle slip value is calculated using cycle slip probe measurements and cycle slip repair is performed. For single-frequency observation data, a non-combined mode is used for cycle slip repair; for multi-frequency linear combination observations, taking dual-frequency observations as an example, two types of linear combination coefficients are required. and Cycle slip measurement and Inverse calculation of cycle slip value at a single frequency point:

[0070] In the formula According to the law of error propagation, the variance corresponding to the cycle slip repair amount at a single frequency point is:

[0071] in, It is related to the combination coefficient and wavelength, and takes the following specific form:

[0072] The formulas for calculating the co-operational term of cycle slip detection under the two combination modes are as follows:

[0073] in, .

[0074] The above formula can be used to back-calculate the phase cycle slip detection value and its variance at a single frequency point. If the detection value exceeds the variance, cycle slip repair is performed. However, a large combination coefficient will amplify the variance corresponding to the phase cycle slip at a single frequency point calculated from the combined observations, making it difficult to accurately determine whether a cycle slip has occurred at a single frequency point. On the other hand, if cycle slips occur simultaneously in both frequency observations and satisfy the coefficient relationship, the detection value cannot be accurately identified. In contrast, non-combined observations do not amplify phase noise and can detect the cycle slip value at a single frequency point. Therefore, for the situation where the performance of the combined detection value deteriorates, non-combined observations are used according to formula (30) for a single frequency point, and the cycle slip value is... Rounding down yields the integer cycle slip repair amount. If this amount satisfies the following formula, the repair amount is considered reliable, and cycle slip repair is performed while increasing the corresponding ambiguity variance:

[0075] in, round The value is rounded to the nearest whole number, and the threshold for the amount of ambiguity correction is set to 0.25 weeks. If the value exceeds this threshold, the correction error is considered too large, and the satellite is treated as a new satellite for ambiguity resolution.

[0076] In summary, the technical solution of this method can be summarized as follows: 1. To address the challenges of frequent cycle slips and difficulties in detection and repair of satellite signals in complex urban environments, a method for cycle slip detection and repair assisted by visual-inertial navigation prior position is proposed. 2. To address the challenge of error accumulation in low-cost visual inertial odometry, a hybrid update method based on long-term and short-term visual features is adopted, which makes full use of the multi-frame co-view relationship of long-term tracking visual features to improve the accuracy of prior position. 3. To address the challenges of significant error accumulation and inaccurate cycle slip detection and repair during long-term satellite signal loss, a multi-frequency linear combination of observations is employed to construct longer-wavelength probes, improving their tolerance to prior position errors. Simultaneously, the low noise of non-combined observations is utilized to achieve single-frequency cycle slip repair, enhancing the reliability of cycle slip repair.

[0077] First, this method avoids the limitations of traditional methods that rely solely on satellite observations. Specifically, it uses the prior position provided by visual inertial odometry (VIO) to construct cycle slip measurements, eliminating the satellite-to-ground geometric distance term in phase observations and avoiding the introduction of observations with large residual errors such as pseudorange. The constructed cycle slip statistics only include prior position errors and phase residual errors. In the short term, the accuracy of the prior position provided by VIO is better than that of pseudorange and pure inertial navigation recursion results, effectively improving the accuracy of cycle slip detection. On the other hand, in complex urban scenarios, pseudorange residual errors include large multipath and non-model-based errors such as non-visual (Non-Line-of-Sight, NLOS) signals, leading to a mismatch between the observed variance and the actual residual error, making it difficult to set an appropriate cycle slip detection threshold. In contrast, VIO, through a rigorous system state model, can provide accurate prior position variance for calculating the cycle slip detection threshold, and is less affected by the observation environment and non-model-based errors, thereby further improving the reliability of cycle slip detection.

[0078] Secondly, this method employs a hybrid update approach based on both short-term and long-term visual features, effectively suppressing the rapid accumulation of low-cost inertial navigation errors to provide more accurate prior positions for cycle slip detection and repair. Specifically, this invention divides visual feature points into short-term and long-term tracking visual features based on the number of tracking frames, and processes them using a hybrid visual update approach: For long-term tracking visual features with a tracking frame count exceeding the number of camera frames within the sliding window, they are added to the state variables as parameters to be estimated through visual feature state augmentation, while simultaneously augmenting their state variable covariance matrix. When a new visual observation value for the feature point is obtained, observation equations for the inertial navigation pose and visual feature points within the window are constructed simultaneously. During filtering and updating, the inertial navigation pose and feature points are optimized simultaneously, fully utilizing the multi-frame co-view relationship of long-term tracking visual features to constrain the rapid accumulation of inertial navigation errors; for short-term tracking visual features with a tracking frame count less than the number of camera frames within the sliding window, the feature point state parameters are eliminated using left null space projection after constructing the visual observation equations, avoiding an increase in the dimensionality of the state variables and the computational load. This strategy can improve the accuracy maintenance capability of visual inertial odometry while ensuring computational efficiency, thereby enhancing the performance of prior position-assisted cycle slip detection.

[0079] Finally, this method leverages the advantage of linearly combining multi-frequency phase observations by employing an adaptive switching strategy between linearly combined and non-combined phase observations, enabling accurate and reliable cycle slip detection and repair. Specifically, this invention linearly combines phase observations at different frequencies to construct cycle slip probes with longer wavelengths, increasing their tolerance to prior position errors and improving cycle slip detection performance under long-term satellite signal lockout conditions. On the other hand, linearly combined observations amplify residual phase errors, and using combined observations to calculate cycle slips at a single frequency further amplifies these residual errors, making it difficult to accurately calculate cycle slips at a single frequency for cycle slip repair. Therefore, after determining that a cycle slip has occurred in a phase observation, if cycle slip repair using combined observations fails, the cycle slip value at a single frequency is calculated using prior position and non-combined observations. The ambiguity parameters of that frequency are then repaired and variance dilation is applied, achieving reliable repair of single-frequency and multi-frequency phase cycle slips.

[0080] Through the aforementioned technological innovations, this method significantly improves the performance of BeiDou satellite carrier phase cycle slip detection and repair in complex urban scenarios, thereby achieving high-precision and high-reliability positioning for low-cost multi-source fusion systems in complex urban environments. Field test results for cycle slip detection and repair in complex urban scenarios show that, using the proposed hybrid long- and short-term visual feature update method, the failure rate of cycle slip detection and repair is reduced by more than 30%, the root mean square deviation of the fractional deviation of the cycle slip repair is improved by 24%, and the detectable duration of a 1-week cycle slip is increased by 2-3 times when all BeiDou satellites are out of lock and when only three satellites are observed, with a positioning error of less than 0.5 m probability of 95.7%. Further, by using linear combination observations for cycle slip detection and repair, it can detect a 1-week cycle slip within 30 seconds when all BeiDou satellites are out of lock, and a 1-week cycle slip within 80 seconds when only three satellites are observed. Overall, the proposed method significantly improves the performance of cycle slip detection and repair under long-term satellite signal loss conditions and has promising application prospects in the field of multi-source fusion navigation and positioning in complex urban scenarios.

[0081] The present invention also provides a visual inertial navigation-assisted BeiDou satellite cycle slip detection and repair device. The visual inertial navigation-assisted BeiDou satellite cycle slip detection and repair device provided by the present invention will be described below. The visual inertial navigation-assisted BeiDou satellite cycle slip detection and repair device described below and the visual inertial navigation-assisted BeiDou satellite cycle slip detection and repair method described above can be referred to in correspondence with each other. Figure 4 This is a structural block diagram of the visual-inertial navigation-assisted BeiDou satellite cycle slip detection and repair device provided by the present invention, as shown in the figure. Figure 4 As shown, the device includes: The preprocessing module 401 is used to acquire the acceleration and angular velocity provided by the inertial measurement unit and predict the prior position of the carrier through dead reckoning. The extraction module 402 is used to acquire the original image, extract and match features from the original image, and divide the visual features of the original image into long-term tracking features and short-term tracking features. The update module 403 is used to construct the observation equation of visual feature points by combining the hybrid update method of long-term and short-term visual features, and to optimize the accuracy of the inertial navigation prior pose and the visual feature state quantity. Repair module 404 is used to construct cycle slip detection measurements using linear and non-linear phase observations, determine whether a cycle slip has occurred, and perform cycle slip repair; the wavelength of the cycle slip detection is greater than the wavelength of the cycle slip probe.

[0082] In operation, the device first employs a preprocessing module 401 to acquire the acceleration and angular velocity provided by the inertial measurement unit (IMU) and predict the prior position of the carrier through dead reckoning. Then, the extraction module 402 acquires the original image, performs feature extraction and matching, and categorizes the visual features of the original image into long-term tracking features and short-term tracking features. Based on this, the update module 403 combines a hybrid update method of long-term and short-term visual features to construct observation equations for visual feature points, optimize the IMU pose and visual feature state variables, and fully utilize the multi-frame co-view relationship of long-term tracking features to constrain the rapid accumulation of errors in the low-cost IMU, thereby improving the accuracy of the prior position. Finally, the repair module 404 constructs cycle slip detection measurements using linearly combined and non-combined phase observations to determine if a cycle slip has occurred and performs cycle slip repair. By constructing cycle slip detection measurements with longer wavelengths, the impact of recursive position errors on cycle slip detection is reduced. Simultaneously, the advantage of low noise from non-combined observations is utilized to achieve accurate repair of cycle slips at a single frequency point. Through the above process, the observation information of low-cost sensors can be fully utilized, significantly improving the accuracy and reliability of cycle slip detection and repair when BeiDou signal quality deteriorates. Furthermore, it can simultaneously handle single-frequency and multi-frequency phase cycle slips of BeiDou satellites, thereby achieving high-precision and high-reliability positioning for a low-cost multi-source fusion system in complex urban scenarios, solving the problems of poor accuracy and reliability in existing related technologies. In addition, this device can also be extended to global satellite navigation systems.

[0083] Figure 5 An example is a schematic diagram of the physical structure of an electronic device, such as... Figure 5 As shown, the electronic device may include: a processor 501, a communication interface 502, a memory 503, and a communication bus 504. The processor 501, communication interface 502, and memory 503 communicate with each other via the communication bus 504. The processor 501 can call logical instructions from the memory 503 to execute a visual-inertial navigation-assisted method for BeiDou satellite cycle slip detection and repair. This method includes: The acceleration and angular velocity provided by the inertial measurement unit are obtained, and the prior position of the vehicle is predicted by dead reckoning. The original image is acquired, and features are extracted and matched. The visual features of the original image are divided into long-term tracking features and short-term tracking features. By combining long-term and short-term visual features in a hybrid update method, an observation equation for visual feature points is constructed to optimize the accuracy of inertial navigation prior pose and visual feature state variables. A longer-wavelength cycle slip detection is constructed using linear and non-linear phase observations to determine whether a cycle slip has occurred and to perform cycle slip repair.

[0084] Furthermore, the logical instructions in the aforementioned memory 503 can be implemented as software functional units and, when sold or used as independent products, can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present invention, or the part that contributes to the prior art, or a part of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods of the various embodiments of the present invention. The aforementioned storage medium includes various media capable of storing program code, such as USB flash drives, portable hard drives, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical disks.

[0085] On the other hand, the present invention also provides a computer program product, which includes a computer program that can be stored on a non-transitory computer-readable storage medium. When the computer program is executed by a processor, the computer is able to execute the visual-inertial navigation-assisted BeiDou satellite cycle slip detection and repair method provided by the above methods, the method comprising: The acceleration and angular velocity provided by the inertial measurement unit are obtained, and the prior position of the vehicle is predicted by dead reckoning. The original image is acquired, and features are extracted and matched. The visual features of the original image are divided into long-term tracking features and short-term tracking features. By combining long-term and short-term visual features in a hybrid update method, an observation equation for visual feature points is constructed to optimize the accuracy of inertial navigation prior pose and visual feature state variables. A longer-wavelength cycle slip detection is constructed using linear and non-linear phase observations to determine whether a cycle slip has occurred and to perform cycle slip repair.

[0086] In another aspect, the present invention also provides a non-transitory computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements a method for visual-inertial navigation-assisted BeiDou satellite cycle slip detection and repair provided by the methods described above, the method comprising: The acceleration and angular velocity provided by the inertial measurement unit are obtained, and the prior position of the vehicle is predicted by dead reckoning. The original image is acquired, and features are extracted and matched. The visual features of the original image are divided into long-term tracking features and short-term tracking features. By combining long-term and short-term visual features in a hybrid update method, an observation equation for visual feature points is constructed to optimize the accuracy of inertial navigation prior pose and visual feature state variables. A longer-wavelength cycle slip detection is constructed using linear and non-linear phase observations to determine whether a cycle slip has occurred and to perform cycle slip repair.

[0087] The device embodiments described above are merely illustrative. The units described as separate components may or may not be physically separate. The components shown as units may or may not be physical units; that is, they may be located in one place or distributed across multiple network units. Some or all of the modules can be selected to achieve the purpose of this embodiment according to actual needs. Those skilled in the art can understand and implement this without any creative effort.

[0088] Through the above description of the embodiments, those skilled in the art can clearly understand that each embodiment can be implemented by means of software plus necessary general-purpose hardware platforms, and of course, it can also be implemented by hardware. Based on this understanding, the above technical solutions, in essence or the part that contributes to the prior art, can be embodied in the form of a software product. This computer software product can be stored in a computer-readable storage medium, such as ROM / RAM, magnetic disk, optical disk, etc., and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute the methods described in the various embodiments or some parts of the embodiments.

[0089] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, and not to limit them; although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features; and these modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of the present invention.

Claims

1. A method for visual-inertial navigation-assisted BeiDou satellite cycle slip detection and repair, characterized in that, include: The acceleration and angular velocity provided by the inertial measurement unit are obtained, and the prior position of the vehicle is predicted by dead reckoning. The original image is acquired, and features are extracted and matched from it. The visual features of the original image are divided into long-term tracking features and short-term tracking features, including: The original image is acquired, features are extracted and matched on the original image, and poor-quality outliers are removed using the random sampling consistency method. The visual feature points of each frame of the image are obtained and the visual features are divided into long-term tracking features and short-term tracking features based on the number of consecutive tracking frames. By combining long-term and short-term visual features in a hybrid update method, an observation equation for visual feature points is constructed to optimize the accuracy of inertial navigation prior pose and visual feature state variables. A longer-wavelength cycle slip detection is constructed using linear and non-linear phase observations to determine whether a cycle slip has occurred and to perform cycle slip repair. Cycle slip detection is constructed using linearly combined and non-combined phase observations to determine whether a cycle slip has occurred, including: The phase observations provided by the base station and the rover are obtained to construct the cycle slip detection measurement. Based on the prior position variance and phase noise, the cycle slip detection threshold is determined. Compare the cycle slip measurement with the cycle slip detection threshold. If the cycle slip measurement exceeds the cycle slip detection threshold, it is determined that a cycle slip has occurred. If the satellite signal loss time is long, use phase observations at different frequencies to make a linear combination to construct a linear combination cycle slip measurement. Compare the linear combination cycle slip measurement with the cycle slip detection threshold. If the linear combination cycle slip measurement is less than the cycle slip detection threshold, it is considered that no cycle slip has occurred, and a double-difference observation equation is constructed for measurement update. If the linear combination cycle slip detection measurement reaches the cycle slip detection threshold, then a cycle slip is considered to have occurred; If a cycle slip occurs, and the decimal part of the cycle slip probe exceeds the preset number of cycles, the deviation is considered too large to be repaired, and the ambiguity of all frequency points of the satellite is reset. If a cycle slip occurs, and the decimal part of the cycle slip probe does not exceed the preset number of cycles, then a linear combination and non-combination mode is used to repair the cycle slip. If the cycle slip of a single frequency point calculated by inverse linear combination is unreliable, then the cycle slip value of a single frequency point is calculated using a non-combination mode, the cycle slip value is corrected to the ambiguity parameter and the state variance is expanded, and then a double-difference observation equation is constructed for quality control and measurement updates.

2. The method for visual-inertial navigation-assisted BeiDou satellite cycle slip detection and repair according to claim 1, characterized in that, Obtaining acceleration and angular velocity from the inertial measurement unit, and predicting the prior position of the vehicle through dead reckoning includes: The acceleration and angular velocity provided by the inertial measurement unit are obtained, and the prior position and its variance at the current moment are determined by dead reckoning.

3. The method for visual-inertial navigation-assisted BeiDou satellite cycle slip detection and repair according to claim 1, characterized in that, By combining long-term and short-term visual feature updates, an observation equation for visual feature points is constructed to optimize the accuracy of the inertial navigation prior pose and the visual feature state variables, including: For long-term tracking features, determine whether the long-term tracking feature is a new visual feature point or an old visual feature point, and update it according to the determination result; For short-term tracking features, after constructing the visual observation equation, the short-term tracking features are eliminated from the parameters to be estimated by using left null space projection. During measurement updates, only the inertial navigation pose is updated, and the visual features are not updated.

4. The method for visual-inertial navigation-assisted BeiDou satellite cycle slip detection and repair according to claim 3, characterized in that, Determining whether the long-term tracked feature is a new visual feature point or an old visual feature point, and updating it based on the determination result, includes: If the feature points of the long-term tracking feature are not added to the state vector, then the long-term tracking feature is a new visual feature point. The new visual feature point is used as the state to be optimized through state augmentation, and the covariance matrix is ​​calculated. The new visual feature point is then added to the state vector. If the feature points of the long-term tracking feature have been added to the state vector, then the long-term tracking feature is the old visual feature point. The new visual observation value of the old visual feature point is used to construct the observation equation about the inertial navigation pose and visual feature points within the window for extended Kalman filter update, and the inertial navigation pose and visual feature points are updated at the same time.

5. A device for visual-inertial navigation-assisted BeiDou satellite cycle slip detection and repair, used to implement the visual-inertial navigation-assisted BeiDou satellite cycle slip detection and repair method according to any one of claims 1-4, characterized in that, include: The preprocessing module is used to acquire the acceleration and angular velocity provided by the inertial measurement unit and predict the prior position of the vehicle through dead reckoning. The extraction module is used to acquire the original image, extract and match features from the original image, and divide the visual features of the original image into long-term tracking features and short-term tracking features. The update module is used to construct the observation equation of visual feature points by combining a hybrid update method of long-term and short-term visual features, and to optimize the accuracy of the inertial navigation prior pose and the visual feature state variables. The repair module is used to construct a longer-wavelength cycle slip probe using linear and non-linear phase observations, determine whether a cycle slip has occurred, and perform cycle slip repair.

6. An electronic device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that, When the processor executes the program, it implements the method for visual inertial navigation-assisted BeiDou satellite cycle slip detection and repair as described in any one of claims 1 to 4.

Images