Combined navigation method based on horizon positioning and imu prior fusion

Abstract

The application provides a kind of combination navigation method based on skyline positioning and IMU prior fusion, belong to navigation technical field, specifically includes: monitoring GNSS signal;The pose information obtained by IMU in short time after GNSS failure is used to construct the prior search space of skyline matching, and the prior space constraint is generated according to the current position and error model;Collect the panoramic image of the environment around the vehicle, process the panoramic image to extract the actual skyline, and match the actual skyline with the preset digital elevation model DEM skyline library under the prior space constraint, output multiple candidate positions, and select the optimal visual positioning result and its confidence from multiple candidate positions based on trajectory consistency score;The optimal visual positioning result is used as pseudo-observation information, and its confidence is used to adaptively adjust the observation noise covariance matrix of the combination navigation filter.The scheme of the application improves the navigation accuracy and robustness.

Description

TECHNICAL FIELD

[0001] The present application relates to the field of navigation technology, and in particular to a combined navigation method based on horizon positioning and IMU prior fusion. BACKGROUND

[0002] At present, although the existing combined navigation system can realize high accuracy and continuity in a short period through the fusion of GNSS and IMU, it still faces many problems in complex or extreme environments. GNSS signals are easily affected by shielding, interference or spoofing attacks, and may be completely invalid in urban canyons, tunnels, forests or wartime environments; IMU will have cumulative errors and drift when lacking external correction signals, resulting in a serious deviation of the positioning result over time, making it difficult to achieve long-term stable navigation. To make up for this deficiency, the introduction of auxiliary positioning methods based on visual information has become a new research direction, but traditional visual positioning is difficult to obtain reliable feature points in the wild natural environment due to the lack of artificial features and texture information. The visual geographic positioning based on horizon improves this problem to some extent, but the existing methods lack robustness under complex weather conditions, and most are static positioning schemes, lacking real-time fusion and dynamic feedback capability with the combined navigation system, making it difficult to meet the continuous high-precision positioning needs of unmanned systems and autonomous platforms in multiple scenarios.

[0003] Therefore, there is an urgent need for a combined navigation method based on horizon positioning and IMU prior fusion with high positioning accuracy and robustness. SUMMARY

[0004] Therefore, the embodiments of the present application provide a combined navigation method based on horizon positioning and IMU prior fusion, which at least partially solves the problem of poor positioning accuracy and robustness in the prior art.

[0005] In a first aspect, the embodiments of the present application provide a combined navigation method based on horizon positioning and IMU prior fusion, comprising:

[0006] Step 1, the system performs IMU and GNSS combined navigation initialization and monitors GNSS signals;

[0007] Step 2, using the pose information obtained by the IMU in a short time after the GNSS failure, a prior search space for horizon matching is constructed, and a prior space constraint is generated according to the current position and error model;

[0008] Step 3, panoramic images of the environment around the vehicle are collected, the panoramic images are processed to extract the actual horizon, and the actual horizon is matched with the preset digital elevation model DEM horizon library under the prior space constraint, a plurality of candidate positions are output, and the optimal visual positioning result and its confidence are selected from the plurality of candidate positions based on trajectory consistency scoring;

[0009] Step 4, taking the optimal visual positioning result as pseudo-observation information, and using the confidence thereof to adaptively adjust the observation noise covariance matrix of the integrated navigation filter, so as to correct the accumulated error of the IMU through the filter, and update the system navigation state.

[0010] According to a specific implementation manner of the embodiment of the present application, the step of generating the prior spatial constraint comprises:

[0011] The latitude and longitude constraint range is determined by using the current position calculated by the IMU and the error model, so as to limit the search area of the DEM skyline library, meanwhile, a directional search window is constructed by using the heading angle calculated by the IMU and the preset directional window threshold, and a directional weight function is constructed according to the directional search window, which is used to apply a penalty weight to a candidate area greatly deviating from the current heading direction in subsequent matching.

[0012] According to a specific implementation manner of the embodiment of the present application, the expression of the latitude and longitude constraint range is:

[0013] ;

[0014] wherein, is the latitude estimation value recorded by the system output by the integrated navigation, is the longitude estimation value recorded by the system output by the integrated navigation, is the standard deviation of the latitude direction in the corresponding state covariance matrix, is the standard deviation of the longitude direction in the corresponding state covariance matrix, and the search scale is dynamically adjusted by the error model.

[0015] According to a specific implementation manner of the embodiment of the present application, the expression of the directional search window is:

[0016] ;

[0017] wherein, is the view angle of each candidate point in the DEM skyline, represents the current heading angle, represents the directional window threshold.

[0018] According to a specific implementation manner of the embodiment of the present application, the step 3 specifically comprises:

[0019] The skyline-DEM matching algorithm is used to extract the visible skyline contour at a given latitude and longitude from the DEM digital elevation model in advance, to construct a skyline reference database, to extract candidate areas from the preset digital elevation model DEM skyline library under the prior spatial constraint, and to compare and output a plurality of candidate positions ranked in the front according to the similarity scores, and to calculate the visual similarity scores for each candidate position.

[0020] A trajectory consistency score function is introduced, which compares the IMU short-time trajectory with the feasible path of each candidate position in the digital elevation model (DEM) skyline library to evaluate the spatial coherence with the inertial trajectory, and a joint scoring model is constructed by fusing the visual matching similarity and the trajectory consistency score;

[0021] The optimal position is selected as the optimal visual positioning result according to the output result of the joint scoring model, and the confidence of the candidate position is estimated based on the score distribution.

[0022] According to a specific implementation manner of the embodiment of the application, the expression of the optimal visual positioning result is as follows:

[0023] ;

[0024] ;

[0025] wherein, is the joint score value of the i th candidate position, indicates the fusion weight of the visual and trajectory constraints, is the visual matching score value of the i th candidate position, is the trajectory consistency score value of the i th candidate position, and thus the optimal matching position is determined, is the time window length for the trajectory consistency score, is the IMU calculated trajectory point, is the backtracking path generated in the digital elevation model (DEM) skyline library with the candidate position as the starting point.

[0026] According to a specific implementation manner of the embodiment of the application, the step 4 specifically comprises:

[0027] Step 4.1, the optimal visual positioning result is introduced into the integrated navigation filter as pseudo-observation information, the error state is updated through the extended Kalman filter to correct the key navigation parameters, wherein the key navigation parameters include the attitude, the speed, the position and the IMU error;

[0028] Step 4.2, the observation noise covariance matrix in the filter is dynamically adjusted according to the confidence, the observation influence is improved when the confidence is high, and the weight is reduced when the confidence is low.

[0029] The integrated navigation scheme based on skyline positioning and IMU prior fusion in this embodiment of the invention includes: Step 1, the system initializes the integrated navigation of IMU and GNSS and monitors GNSS signals; Step 2, the system constructs a prior search space for skyline matching using the pose information calculated by the IMU in a short time after GNSS failure, and generates prior space constraints based on the current position and error model; Step 3, the system acquires panoramic images of the environment around the vehicle, processes the panoramic images to extract the actual skyline, and matches the actual skyline with a preset digital elevation model (DEM) skyline library under the prior space constraints, outputs multiple candidate positions, and selects the optimal visual positioning result and its confidence level from the multiple candidate positions based on the trajectory consistency score; Step 4, the optimal visual positioning result is used as pseudo-observation information, and its confidence level is used to adaptively adjust the observation noise covariance matrix of the integrated navigation filter, thereby correcting the cumulative error of the IMU through the filter and updating the system navigation status.

[0030] The beneficial effects of this invention are as follows: By introducing prior pose information calculated by IMU to constrain the visual positioning process, a two-way fusion mechanism is constructed that integrates IMU prior-assisted visual positioning with visual pseudo-observation feedback correction for combined navigation. This method fully combines skyline visual features and DEM terrain information, actively narrowing the search space before visual positioning to improve matching efficiency and accuracy, and providing feedback correction for inertial navigation errors after visual positioning to achieve a better closed loop. Attached Figure Description

[0031] To more clearly illustrate the technical solutions of the embodiments of the present invention, the drawings used in the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0032] Figure 1 A flowchart illustrating a combined navigation method based on skyline positioning and IMU prior fusion provided in an embodiment of the present invention;

[0033] Figure 2 This is a schematic diagram illustrating the specific implementation process of a combined navigation method based on skyline positioning and IMU prior fusion provided in an embodiment of the present invention;

[0034] Figure 3 A candidate selection map based on IMU position and heading prior is provided for an embodiment of the present invention;

[0035] Figure 4 This is a schematic diagram of trajectory consistency scoring provided in an embodiment of the present invention. Detailed Implementation

[0036] The embodiments of the present invention will now be described in detail with reference to the accompanying drawings.

[0037] The following specific examples illustrate the implementation of the present invention. Those skilled in the art can easily understand other advantages and effects of the present invention from the content disclosed in this specification. Obviously, the described embodiments are only a part of the embodiments of the present invention, and not all of them. The present invention can also be implemented or applied through other different specific embodiments, and the details in this specification can also be modified or changed based on different viewpoints and applications without departing from the spirit of the present invention. It should be noted that, in the absence of conflict, the following embodiments and features in the embodiments can be combined with each other. All other embodiments obtained by those skilled in the art based on the embodiments of the present invention without creative effort are within the scope of protection of the present invention.

[0038] It should be noted that various aspects of embodiments within the scope of the appended claims are described below. It will be apparent that the aspects described herein can be embodied in a wide variety of forms, and any particular structure and / or function described herein is merely illustrative. Based on this invention, those skilled in the art will understand that one aspect described herein can be implemented independently of any other aspect, and two or more of these aspects can be combined in various ways. For example, any number of aspects set forth herein can be used to implement the device and / or practice the method. Additionally, this device and / or method can be implemented using structures and / or functionalities other than one or more of the aspects set forth herein.

[0039] It should also be noted that the illustrations provided in the following embodiments are only schematic representations of the basic concept of the present invention. The illustrations only show the components related to the present invention and are not drawn according to the actual number, shape and size of the components in the actual implementation. In the actual implementation, the form, quantity and proportion of each component can be arbitrarily changed, and the layout of the components may also be more complex.

[0040] Furthermore, specific details are provided in the following description to facilitate a thorough understanding of the examples. However, those skilled in the art will understand that the described aspects can be practiced without these specific details.

[0041] With the development of modern navigation systems, integrated navigation systems have gradually become a core technology in applications such as unmanned systems, autonomous vehicles, and military equipment. Integrated navigation typically refers to a multi-source fusion navigation method that combines the Global Positioning System (GNSS) with an Inertial Navigation System (IMU). GNSS provides absolute position reference, while the IMU calculates position through acceleration and angular velocity measurements. The two complement each other, achieving high accuracy and continuity in the short term.

[0042] However, integrated navigation systems still suffer from the following problems: GNSS signals are susceptible to obstruction, interference, or spoofing attacks, and are prone to failure in urban canyons, tunnels, forests, underground spaces, or wartime environments; IMUs accumulate errors and drift in the absence of external correction signals, causing positioning results to deviate significantly over time. Therefore, existing integrated navigation systems struggle to maintain stable and reliable positioning performance during long-term operation or in extreme environments.

[0043] To compensate for the shortcomings of integrated navigation systems in specific scenarios, visual information is introduced as an auxiliary source. Image features are extracted from the environment and used to participate in navigation state estimation; this approach is called visual geolocation. This method does not rely on external base stations, networks, or satellites, and has high environmental adaptability, making it particularly suitable for scenarios with limited communication or GNSS failure.

[0044] In natural outdoor scenes, traditional visual positioning methods struggle to acquire stable key features due to the lack of man-made structures and sparse texture information. To obtain more suitable positioning information sources, Petovello et al. proposed a concept based on visible skylines at specific points. The skyline, as the boundary between the sky and non-sky areas in an image, has advantages such as stability, strong uniqueness, and minimal impact from environmental occlusion. Ramalingam et al. proposed a method using an omnidirectional camera and a textureless 3D city model, utilizing the skyline extracted from panoramic images for geolocation. Experimental results show that this method outperforms GNSS measurements in specific environments. Petovello et al. further proposed using a low-cost infrared camera to isolate the skyline from captured infrared images and match it with the skyline generated from a 3D city model for positioning in urban environments. However, the above methods lack robustness in complex weather conditions and are static positioning schemes, lacking real-time interaction and fusion capabilities with integrated navigation systems, and unable to provide dynamic feedback or correction to the navigation status.

[0045] This invention provides a combined navigation method based on skyline positioning and IMU prior fusion, which can be applied to the GNSS and IMU combined navigation process in navigation scenarios.

[0046] See Figure 1 This is a flowchart illustrating a combined navigation method based on skyline positioning and IMU prior fusion, provided by an embodiment of the present invention. Figure 1 and Figure 2 As shown, the method mainly includes the following steps:

[0047] Step 1: The system initializes the IMU and GNSS integrated navigation and monitors GNSS signals;

[0048] In practical implementation, the integrated navigation system can adopt a loosely coupled fusion architecture of IMU and GNSS. During the system startup phase, the initial absolute position is obtained through GNSS, and attitude and velocity initialization is performed by combining the angular velocity and acceleration output by the IMU. The system constructs a main loop based on extended Kalman filtering, relies on the IMU to achieve high-frequency state prediction, and introduces position observations to complete periodic corrections when GNSS is available.

[0049] During the operation of integrated navigation, the system continuously monitors GNSS signal quality parameters in real time, sets multiple judgment thresholds and sliding decision windows. If a decline in GNSS data quality or complete loss is detected within multiple consecutive cycles, the system will automatically identify that it is currently in a GNSS unavailable state and trigger the subsequent visual-assisted positioning module to prepare for IMU-guided skyline matching positioning.

[0050] Step 2: Construct a prior search space for skyline matching using the pose information obtained by the IMU shortly after GNSS failure, and generate prior space constraints based on the current position and error model.

[0051] In practice, the system uses the position estimation results obtained from short-time IMU calculations and combines them with a dynamic error model to construct a matching latitude and longitude constraint area to limit the spatial range of visual matching. This is in response to GNSS failure. The system records the position estimate output by the integrated navigation system. And extract the diagonal elements of the corresponding covariance matrix. , To ensure that the candidate region covers the probability of the true location, the system establishes a non-equidistant elliptical search region centered on this point, with a spatial range of:

[0052] ;

[0053] Combined with rasterized DEM region indexing, spatial-level constraints can be achieved. When IMU accuracy is low, the elliptical region expands, significantly increasing the computational cost of matching. This strategy also adaptively adjusts the dynamic error inflation coefficient. , Ensure robustness.

[0054] Subsequently, the system constructs a directional search window based on the heading angle information output by the current integrated navigation system, and uses the visible skyline of each candidate point in the DEM to generate directions for angle difference filtering, enhancing the consistency of direction matching. This is achieved by obtaining the current heading angle. And set the orientation window threshold. Construct forward perspective constraints:

[0055] ;

[0056] in, For the set of directional constraints, Generate a view of the skyline in the DEM for each candidate point. If the candidate point's orientation is not specified... Within this range, regions are preemptively removed to avoid invalid calculations. This method not only reduces the burden of DEM extraction but also enhances the directional consistency of candidate regions. The resulting candidate selection map based on IMU position and heading priors is shown below. Figure 3 As shown.

[0057] The system further utilizes a directional confidence modeling mechanism, introducing a Gaussian directional weight function:

[0058] ;

[0059] The closer the direction is to the heading The higher the weight, the better. This function will be used as a priori scoring item when scoring candidate regions, and will be used for subsequent fusion with visual matching scores.

[0060] Finally, the system constructs a joint direction-space prior scoring function to assign multi-dimensional prior weights to DEM candidate points, and fuses it with the image similarity matching model during the candidate scoring stage to achieve dynamic candidate ranking under multiple constraints. In the matching stage, let the visual similarity score be... directional weights are The spatial boundary weights (whether they are within the error ellipse) are: The final comprehensive prior scoring function is defined as follows:

[0061] ;

[0062] in, Let be the comprehensive prior score for the i-th candidate position. A binary function (or further extended to a decaying weight function) represents the points. Whether it is within the prior latitude and longitude region. The final system selects scores before... Candidate points It then enters the trajectory scoring stage and merges with subsequent states.

[0063] The above mechanism unifies the IMU prior position, heading, and visual similarity into a scoring quantification expression, realizes multi-dimensional filtering and fusion before visual matching, and constructs a visual decoupling mechanism guided by physical prior constraints, which greatly reduces the probability of false matching.

[0064] Step 3: Acquire panoramic images of the environment around the vehicle, process the panoramic images to extract the actual skyline, and match the actual skyline with the preset digital elevation model (DEM) skyline library under prior spatial constraints to output multiple candidate locations. Then, select the optimal visual positioning result and its confidence level from the multiple candidate locations based on the trajectory consistency score.

[0065] In practice, when GNSS is unavailable, the system parks at a fixed point in an open area and uses a vehicle-mounted gimbal to control the camera to capture multi-view images, obtaining a 360° image sequence. The images are uniformly projected onto a single coordinate system using cylindrical projection, and stitched together based on feature detection and matching methods. RANSAC is then used to remove mismatched points to generate a panoramic image. Overlapping areas are eliminated through weighted fusion to remove seams, and the recorded gimbal attitude information is used for calibration to ensure the images conform to a horizontal reference.

[0066] The corrected panoramic image is input into a skyline segmentation model, which is built on a convolutional neural network and outputs probability maps of sky and non-sky regions. After obtaining a binary mask through thresholding, the system uses a column-by-column extremum extraction method to determine the transition points between sky and non-sky regions, thereby forming a skyline point set, which is then converted into a corresponding angle set to finally obtain the image skyline contour. .

[0067] Then, under the latitude-longitude-direction constraints constructed in step 2, the system extracts a set of candidate locations that meet the conditions from the DEM. And generate a simulated skyline at each point. Calculate the Euclidean distance matching error between the image skyline and the simulated skyline.

[0068] The system incorporates multiple physical prior constraints to score and filter candidate matching results, avoiding false matches caused by relying solely on skyline shape similarity. Directional weights are also introduced. Spatial confidence weights Construct a joint visual matching scoring model:

[0069] ;

[0070] in, Gaussian score representing the consistency between the candidate position viewpoint and the IMU heading; This represents the spatial confidence of a candidate point within the error ellipse, where the elliptic error distance function is defined as:

[0071] ;

[0072] This scoring model integrates three sources: image appearance matching, orientation consistency, and prior uncertainty in navigation estimation. Compared to single visual matching, this fusion mechanism significantly improves the interpretability and reliability of localization results, and is particularly robust under conditions such as lack of texture and abrupt changes in lighting.

[0073] The system selects the top K candidate points by score to proceed to the next stage. To further eliminate mismatches and enhance path continuity, the system introduces a trajectory consistency scoring mechanism, which verifies the matching results based on spatial and temporal information. The core idea is that if a candidate point's historical trajectory has a continuous feasible path in the DEM, and this path has high consistency with the short-term inference trajectory from the IMU, then the location reliability of that candidate point should also be higher. The trajectory consistency scoring diagram is shown below. Figure 4 As shown.

[0074] Therefore, the system constructs the following trajectory consistency score:

[0075] ;

[0076] in To calculate trajectory points for the IMU, For candidate positions The reverse path is generated from the DEM, starting from the given location. This score measures the geometric coherence of the path and is of significant value in areas of terrain discontinuity, such as urban edges and near ridgelines.

[0077] Finally, a joint scoring model is constructed:

[0078] ;

[0079] in This represents the fusion weight of visual and trajectory constraints. After fusion, it can effectively alleviate the single-point interference caused by visual mismatches to the navigation results, while retaining the path continuity advantage brought by IMU prediction.

[0080] Based on this, the system determines the optimal matching position:

[0081] ;

[0082] To further support confidence modeling in integrated navigation filtering, the system also estimates the confidence level of the current visual positioning result. The indicators reflect the consistency of the candidate rankings.

[0083] Step 4: The optimal visual positioning result is used as pseudo-observation information, and its confidence is used to adaptively adjust the observation noise covariance matrix of the integrated navigation filter. Then, the cumulative error of the IMU is corrected by the filter, and the system navigation state is updated.

[0084] In practice, the system obtains the optimal positioning result in step 3. Simultaneously, the location reliability index corresponding to the matching result is output. To avoid low-confidence observations misleading the navigation status, the system first sets a confidence fusion threshold. Based on this, three fusion strategies are defined:

[0085] when At this time, the system treats the positioning result as a high-confidence pseudo-observation and uses it to fully participate in the position observation correction of the integrated navigation filter.

[0086] when In this case, the system will only use the pseudo-observation for weak constraint updates in the location dimension, or as a soft condition for other judgment modules to refer to.

[0087] when If the value is significantly lower than the threshold, the system abandons the current fusion and skips the current pseudo-observation to maintain system stability and prevent erroneous information from being introduced into the main state loop.

[0088] After determining the fusion strategy, the system further... The observation noise covariance matrix used in the dynamic adjustment filter ensures that when the visual matching quality is high, the observation noise is lower and the fusion weight is higher; while when the quality is poor, the observation impact is automatically reduced.

[0089] The visual positioning results are then input into the integrated navigation filter as pseudo-observation vectors, and a state correction term is constructed based on the quality perception mechanism.

[0090] ;

[0091] in, The current calculated state gain of the filter. This is the observation matrix corresponding to the pseudo-observation. This represents the current prediction state. The correction term will adjust the confidence level. By integrating into the state fusion process, a dynamic mapping relationship between observation reliability and state adjustment magnitude is realized.

[0092] After completing the state correction term calculation, the optimal positioning result obtained in step 3 will be used. The position observation channel of the Extended Kalman Filter (EKF) is introduced as a pseudo-observation vector. By fusing the residual between the state prediction value and the pseudo-observation, the current state is updated. The corrected state not only includes the position update, but also links the correction of velocity, attitude and IMU error sub-states, realizing multi-dimensional joint estimation.

[0093] Finally, the system outputs the fused state vector, including dynamically corrected three-dimensional position, velocity, attitude and their covariance, for subsequent use by the navigation module. The system then continues to combine the navigation main loop and GNSS monitoring tasks with the fused and updated state, realizing deep fusion and closed-loop feedback of visual pseudo-observations and inertial navigation calculations, thereby completing the deep fusion and feedback closed loop of visual positioning information and inertial navigation system.

[0094] The integrated navigation method based on skyline positioning and IMU prior fusion provided in this embodiment constrains the visual positioning process by introducing prior pose information calculated by the IMU, and constructs a two-way fusion mechanism that integrates IMU prior-assisted visual positioning and visual pseudo-observation feedback correction for integrated navigation. This method fully combines skyline visual features and DEM terrain information, actively narrowing the search space before visual positioning to improve matching efficiency and accuracy, and providing feedback correction for inertial navigation errors after visual positioning to achieve a better closed loop.

[0095] Compared with the traditional IMU-GNSS integrated navigation method that relies entirely on IMU calculation during GNSS signal failure, this invention introduces a feedback correction mechanism based on visual-assisted positioning at key nodes. At the same time, it uses IMU output to restrict the visual matching area in reverse, achieving high-precision, low-latency two-way coupling, which significantly enhances the system's positioning reliability and error control capability.

[0096] The periodic correction mechanism of "parking and shooting - prior constraints - visual matching - state feedback" constructed in this invention has good adaptability and deployment flexibility. When GNSS is unavailable, the vehicle can collect images at fixed points in a suitable area, and combine IMU constraints to quickly match the DEM to simulate the skyline and invert the current position, significantly reducing the dependence on continuous GNSS.

[0097] This invention is particularly suitable for mountainous and canyon areas, filling a critical gap in addressing the decline in navigation accuracy during GNSS failure phases, and has significant engineering application value. Furthermore, this method possesses good system compatibility and ease of deployment, and can be loosely integrated with existing combined navigation systems, demonstrating broad prospects for engineering applications.

[0098] It should be understood that various parts of the present invention can be implemented in hardware, software, firmware, or a combination thereof.

[0099] The above description is merely a specific embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any variations or substitutions that can be easily conceived by those skilled in the art within the technical scope disclosed in the present invention should be included within the scope of protection of the present invention. Therefore, the scope of protection of the present invention should be determined by the scope of the claims.

Claims

1. A method for integrated navigation based on horizon positioning and IMU prior fusion, characterized in that, The method comprises the following steps: Step 1, the system performs IMU and GNSS combined navigation initialization, and monitors GNSS signals; Step 2, a prior search space for skyline matching is constructed by using the position information obtained by IMU after GNSS failure, and a prior space constraint is generated according to the current position and error model; The step of generating the prior space constraint comprises: The current position calculated by IMU and the error model are used to determine the latitude and longitude constraint range, so as to limit the search area of the digital elevation model DEM skyline library, and the heading angle calculated by IMU and the preset direction window threshold are used to construct a directional search window, and a directional weight function is constructed according to the directional search window, which is used to apply a penalty weight to the candidate area deviating greatly from the current direction in subsequent matching; Step 3, panoramic images of the environment around the vehicle are collected, the panoramic images are processed to extract the actual skyline, and the actual skyline is matched with the preset digital elevation model DEM skyline library under the prior space constraint, a plurality of candidate positions are output, and the optimal visual positioning result and its confidence are selected from the plurality of candidate positions based on trajectory consistency score; The step 3 specifically comprises: A skyline-DEM matching algorithm is used to extract the visible skyline contour under the given latitude and longitude from the digital elevation model DEM in advance, a skyline reference database is constructed, candidate areas are extracted from the preset digital elevation model DEM skyline library under the prior space constraint, and a plurality of candidate positions with high similarity score are output, and the visual similarity score of each candidate position is calculated; A trajectory consistency score function is introduced, the IMU short trajectory is compared with the feasible path of each candidate position in the digital elevation model DEM skyline library, the continuity of the trajectory in space is evaluated, a joint score model is constructed by fusing the visual matching similarity and the trajectory consistency score; The optimal position is selected as the optimal visual positioning result according to the output result of the joint score model, and the confidence of the candidate position is estimated based on the score distribution of the candidate position; Step 4, the optimal visual positioning result is used as pseudo-observation information, and the confidence thereof is used to adaptively adjust the observation noise covariance matrix of the combined navigation filter, so as to correct the cumulative error of the IMU through the filter, and update the navigation state of the system.

2. The method of claim 1, wherein, The expression of the latitude and longitude constraint range is: wherein, a latitude estimate value is recorded by the system for the combined navigation output, a longitude estimate value is recorded by the system for the combined navigation output, a standard deviation in the latitude direction in the corresponding state covariance matrix, a standard deviation in the longitude direction in the corresponding state covariance matrix, the search scale is dynamically adjusted by the error model.

3. The method of claim 2, wherein, The expression of the directional search window is: wherein, generating a view perspective of the horizon in the DEM for each candidate point, denotes the current heading angle, denotes the direction window threshold.

4. The method of claim 3, wherein, The expression of the optimal visual positioning result is: wherein, is a joint score value for the i-th candidate position, denotes a fusion weight of vision and trajectory constraints, is a vision matching score value for the i-th candidate position, is a trajectory consistency score value for the i-th candidate position, whereby the optimal matching position is determined, is a time window length for the trajectory consistency score, is an IMU resolved trajectory point, is a candidate position is a backtracking path generated in a digital elevation model (DEM) skyline library from the origin.

5. The method of claim 4, wherein, The step 4 specifically comprises: Step 4.1, the optimal visual positioning result is introduced into the combined navigation filter as pseudo-observation information, the error state is updated through extended Kalman filtering, and the key navigation parameters are corrected, wherein the key navigation parameters comprise attitude, speed, position and IMU error; Step 4.2, the observation noise covariance matrix in the filter is dynamically adjusted according to the confidence, the observation influence is improved when the confidence is high, and the weight is reduced when the confidence is low.

Images