Environment-adaptive high-precision Beidou vision SLAM (Simultaneous Localization and Mapping) cooperative localization optimization method

Abstract

The invention discloses an environment-adaptive high-precision Beidou visual SLAM (Simultaneous Localization and Mapping) cooperative localization optimization method. The method comprises the following steps: step 1, acquiring environment characteristic data, visual SLAM data and Beidou system data; 2, constructing an environment self-adaptive model, wherein the environment self-adaptive model adjusts the weight distribution of the visual SLAM and the Beidou system according to the environment characteristic data; and step 3, performing data fusion on the visual SLAM data and the Beidou system data according to respective weight distribution to form a unified positioning result. Through the environment adaptive model, the Kalman filtering algorithm is improved, and the mechanism is adjusted in real time, so that the cooperative positioning precision and robustness of the visual SLAM and Beidou system are effectively improved.

Description

Technical Field

[0001] The present invention relates to the fields of autonomous driving, unmanned aerial vehicle navigation and robot positioning technology, and in particular to an environment-adaptive high-precision Beidou vision SLAM collaborative positioning optimization method. Background Art

[0002] With the development of autonomous driving technology and drone technology, real-time, high-precision positioning and mapping have become one of the key technologies. Existing positioning systems mainly include visual SLAM (Simultaneous Localization and Mapping), GPS, Beidou Satellite Navigation System (hereinafter referred to as Beidou System), etc. Visual SLAM (Simultaneous Localization and Mapping) technology is a technology for simultaneous positioning and mapping based on visual information. Visual SLAM obtains image data of the environment through a camera and uses image features for positioning and map creation. SLAM technology is widely used in robots, drones, augmented reality and other fields. Current visual SLAM technologies include feature point matching (such as ORB-SLAM), direct methods (such as DSO) and graph optimization methods (such as G2O), which show different advantages and disadvantages in different environments. Beidou Satellite Navigation System is a global satellite navigation system with the advantages of global coverage, high-precision positioning, and strong real-time performance. The Beidou system can provide meter-level positioning accuracy and gradually develop towards sub-meter-level accuracy. The high reliability and global coverage of its navigation signals make it the first choice for many high-precision applications. SLAM technology uses visual information of the environment to achieve map construction and positioning, while the Beidou system provides global positioning services. Combining the Beidou system and visual SLAM technology can bring out the advantages of both. The Beidou system provides coarse positioning information on a global scale, while visual SLAM provides high-precision local environmental information. By integrating these two technologies, better results can be achieved in both wide-area positioning and local refined positioning. In recent years, many researchers have been committed to combining the Beidou system with visual SLAM technology to improve the overall positioning accuracy and robustness of the system.

[0003] However, the existing visual SLAM technology has limitations in performance in different environments. In particular, the performance of visual SLAM may be greatly reduced in environments with drastic lighting changes, single textures, or frequent dynamic obstacles. For example, in low-light environments, the degradation of image quality may make it difficult to extract and match feature points, thereby affecting positioning accuracy. In dynamic environments, moving objects may cause mismatching of feature points, which in turn affects the stability and accuracy of SLAM. In some applications, the Beidou system usually provides positioning services alone. The Beidou system is positioned through satellite signals and can provide positioning information worldwide. However, the Beidou system may experience signal obstruction or interference in some special environments (such as urban canyons, underground parking lots, etc.), resulting in reduced positioning accuracy. In addition, the positioning accuracy of the Beidou system may vary in different regions, affecting its performance in high-precision applications. These systems each have certain limitations. Therefore, visual SLAM is combined with a satellite navigation system (such as the Beidou system) to improve positioning accuracy through multi-sensor fusion, thereby improving the robustness of the system.

[0004] Traditional methods of combining Beidou with visual SLAM usually adopt simple data fusion strategies. For example, the positioning information provided by the Beidou system is used as the initial position or global reference of visual SLAM, and the two are weightedly fused through methods such as Kalman filtering. Although these methods have improved the positioning accuracy to a certain extent, there are still problems of insufficient fusion accuracy and poor real-time performance in complex environments. In addition, these methods often do not fully consider the impact of environmental changes on system performance. In complex environments, the robustness and accuracy of the SLAM algorithm will be affected to a certain extent, such as lighting changes, dynamic obstacles, etc. Moreover, there are usually accuracy differences in the data fusion of the Beidou system and visual SLAM, which will also lead to positioning errors. In addition, the calculation of high-precision positioning algorithms is complex, and real-time performance is difficult to guarantee. Summary of the invention

[0005] The purpose of the present invention is to overcome the defects of the Beidou system and visual SLAM in the prior art in that the fusion accuracy is insufficient and the real-time performance is poor when performing weighted fusion, resulting in large positioning errors and complex calculations, and to provide an environment-adaptive high-precision Beidou visual SLAM collaborative positioning optimization method.

[0006] The purpose of the present invention is achieved through the following technical solutions: An environment-adaptive high-precision Beidou visual SLAM collaborative positioning optimization method comprises the following steps: Step 1, collect environmental feature data, visual SLAM data and Beidou system data; Step 2: construct an environment adaptive model, which adjusts the weight distribution of visual SLAM and Beidou system according to environmental feature data; Step 3: Fusion the visual SLAM data and BeiDou system data according to their respective weight distribution to form a unified positioning result.

[0007] Preferably, in step 1, the collected environmental feature data includes current light intensity, feature point distribution of visual SLAM and Beidou signal quality.

[0008] Preferably, the step 2 is specifically as follows: Constructing an environment adaptive model, the environment adaptive model analyzes the environment feature data to determine the characteristics of the current environment, wherein the characteristics of the current environment are: determining whether the current environment is well-lit or insufficiently lit according to the light intensity, judging the richness of the distribution of feature points in the current environment, and the strength and stability of the Beidou signal; According to the characteristics of the current environment, adjust the weight distribution of visual SLAM and Beidou system: in an environment with good lighting and rich distribution of feature points, increase the weight of visual SLAM; in an environment with insufficient lighting and insufficient distribution of feature points, reduce the weight of visual SLAM; in an environment with strong and stable Beidou signals, increase the weight of the Beidou system; in an environment with weak and unstable Beidou signals, reduce the weight of the Beidou system.

[0009] As a preferred method, the environment-adaptive high-precision Beidou visual SLAM collaborative positioning optimization method also optimizes the environment-adaptive model, and improves the adaptability and accuracy of the model through an adaptive learning mechanism, specifically: Collect historical environmental characteristic data and pre-process the historical environmental characteristic data to remove noise and outliers; Build a prediction model and divide the historical environmental feature data into a training set and a test set. The training set is used to train the model, and the test set is used to evaluate the performance of the model. The training set is used to train the prediction model. Based on historical environmental feature data, the prediction model is used to obtain the predicted positioning result, the predicted positioning result is compared with the test set, the error of the predicted positioning result is calculated, the model parameters are adjusted to reduce the error, and the performance of the prediction model is improved.

[0010] Preferably, if the similarity between the collected environmental feature data and the historical environmental feature data is higher than a set threshold, the positioning result corresponding to the historical environmental feature data is matched with the current collected environmental feature data.

[0011] Preferably, the environment-adaptive high-precision Beidou visual SLAM collaborative positioning optimization method also includes step 4, and the positioning result is also monitored and adjusted in real time, specifically: Real-time monitoring of positioning accuracy and environmental changes; According to the real-time monitoring results, the parameters of the environment adaptive model are dynamically adjusted to optimize the weight distribution of the visual SLAM and passive systems; The positioning error is corrected and data fusion is performed again according to the weight distribution to form a new unified positioning result.

[0012] Preferably, in step 3, the visual SLAM data and the Beidou system data are fused according to their respective weight distributions through a Kalman filter, specifically: Step 3a, preprocessing the visual SLAM data and the Beidou system data so that the visual SLAM data and the Beidou system data have the same timestamp and coordinate system, and performing weighted processing on the visual SLAM data and the Beidou system data according to the weight distribution determined by the environment adaptive model; Step 3b, initializing the state vector and covariance matrix of the Kalman filter, the state vector includes position and velocity positioning parameters, and the covariance matrix is ​​used to represent the uncertainty of state estimation; Step 3c, using the time update equation of the Kalman filter, predicting the state estimate and covariance matrix at the current moment based on the state estimate and process noise covariance matrix at the previous moment; Step 3d, input the weighted visual SLAM data and Beidou system data into the Kalman filter as observation values, use the measurement update equation of the Kalman filter, and update the state estimate and covariance matrix according to the observation values ​​and the observation noise covariance matrix; Step 3e, repeat steps 3c to 3d until the required positioning accuracy or number of iterations is achieved, and output the final updated state estimate as a unified positioning result.

[0013] The beneficial effects of the present invention are as follows: the present invention effectively improves the collaborative positioning accuracy and robustness of the visual SLAM and Beidou system through the environmental adaptive model, the improvement of the Kalman filter algorithm and the real-time adjustment mechanism. Under various environmental conditions, the positioning accuracy is significantly improved compared with the traditional method. At the same time, the system can adapt to different environmental changes, such as illumination changes, sparse feature points, etc., and improves the performance in complex environments. By adaptively adjusting the weights, the resource utilization of the visual SLAM and Beidou system is optimized, and the overall efficiency of the system is improved. BRIEF DESCRIPTION OF THE DRAWINGS

[0014] Figure 1 It is a flow chart of the present invention. DETAILED DESCRIPTION

[0015] The present invention is further described below in conjunction with the accompanying drawings and embodiments.

[0016] Example: An environment-adaptive high-precision Beidou visual SLAM collaborative positioning optimization method, such as Figure 1 As shown, the following steps are included: Step 1, collect environmental feature data, visual SLAM data and Beidou system data; Step 2: construct an environment adaptive model, which adjusts the weight distribution of visual SLAM and Beidou system according to environmental feature data; Step 3: Fusion the visual SLAM data and BeiDou system data according to their respective weight distribution to form a unified positioning result.

[0017] In the step 1, the collected environmental feature data include the current light intensity, the feature point distribution of the visual SLAM and the Beidou signal quality.

[0018] The step 2 is specifically as follows: Constructing an environment adaptive model, the environment adaptive model analyzes the environment feature data to determine the characteristics of the current environment, wherein the characteristics of the current environment are: determining whether the current environment is well-lit or insufficiently lit according to the light intensity, judging the richness of the distribution of feature points in the current environment, and the strength and stability of the Beidou signal; According to the characteristics of the current environment, adjust the weight distribution of visual SLAM and Beidou system: in an environment with good lighting and rich distribution of feature points, increase the weight of visual SLAM; in an environment with insufficient lighting and insufficient distribution of feature points, reduce the weight of visual SLAM; in an environment with strong and stable Beidou signals, increase the weight of the Beidou system; in an environment with weak and unstable Beidou signals, reduce the weight of the Beidou system.

[0019] In this embodiment, the weight values ​​of the Beidou system and the visual SLAM in different environmental feature data can be preset, which can be specifically as follows: Environment 1: Good lighting and rich distribution of feature points.

[0020] Environmental characteristic data: Light intensity: 1000 lux (strong light) Feature point distribution density: 50 feature points per square meter (high density) Beidou signal quality: signal strength 80% (strong), stability 95% (high) Weight distribution: Visual SLAM weight: 0.7 (high) In an environment with good lighting and rich distribution of feature points, visual SLAM can accurately extract and match feature points, so it is given a higher weight. Although the Beidou system has a strong and stable signal, its contribution is relatively small in this local fine positioning environment.

[0021] Through experiments, we can see that the visual SLAM positioning results are: longitude X1, latitude Y1, error 0.1 meter, and the Beidou system positioning results are: longitude X2, latitude Y2, error 1.0 meter.

[0022] The positioning results after fusion are: longitude (X10.7+X20.3), latitude (Y10.7+Y20.3), with an error of less than 0.2 meters.

[0023] Environment 2: Insufficient lighting and sparse distribution of feature points.

[0024] Environmental characteristic data: Light intensity: 50lux (weak light) Feature point distribution density: 10 feature points per square meter (low density) Beidou signal quality: signal strength 60% (medium), stability 80% (medium) Weight distribution: Visual SLAM weight: 0.3 (low) Beidou system weight: 0.7 (high) In an environment with insufficient lighting and sparsely distributed feature points, the positioning performance of visual SLAM will drop significantly, so it is given a lower weight. Although the signal strength and stability of the Beidou system are not as good as those in environment 1, its contribution is relatively large in this global positioning environment.

[0025] Through experiments, we can see that the visual SLAM positioning results are: longitude X3, latitude Y3, error 2.0 meters, and the Beidou system positioning results are: longitude X4, latitude Y4, error 1.5 meters.

[0026] The positioning results after fusion are: longitude (X30.3+X40.7), latitude (Y30.3+Y40.7), with an error of less than 1.8 meters.

[0027] Environment three: Beidou signal is strong and stable.

[0028] Environmental characteristic data: Light intensity: 500 lux (medium light) Feature point distribution density: 30 feature points per square meter (medium density) Beidou signal quality: signal strength 90% (strong), stability 98% (high) Weight distribution: Visual SLAM weight: 0.5 (medium) BeiDou system weight: 0.5 (medium, but tends to be high) Under medium illumination and feature point distribution density, the positioning performance of visual SLAM and Beidou system is relatively good. However, considering that Beidou signal is strong and stable, it is given a slightly higher weight to make full use of its global positioning advantage.

[0029] Through experiments, we can see that the visual SLAM positioning result is: longitude X5, latitude Y5, error 0.5 meters, and the Beidou system positioning result is: longitude X6, latitude Y6, error 0.8 meters.

[0030] The positioning results after fusion are: longitude (X50.5+X60.5), latitude (Y50.5+Y60.5), with an error of less than 0.65 meters.

[0031] Environment 4: Beidou signal is weak and unstable.

[0032] Environmental characteristic data: Light intensity: 800 lux (strong light, but may be affected by cloud cover) Feature point distribution density: 40 feature points per square meter (high density, but may be affected by shadows) Beidou signal quality: signal strength 40% (weak), stability 70% (low) Weight distribution: Visual SLAM weight: 0.6 (high) Beidou system weight: 0.4 (low) The environment-adaptive high-precision Beidou visual SLAM collaborative positioning optimization method also optimizes the environment-adaptive model and improves the adaptability and accuracy of the model through an adaptive learning mechanism. Specifically: Collect historical environmental characteristic data and pre-process the historical environmental characteristic data to remove noise and outliers; Build a prediction model and divide the historical environmental feature data into a training set and a test set. The training set is used to train the model, and the test set is used to evaluate the performance of the model. The training set is used to train the prediction model. Based on historical environmental feature data, the prediction model is used to obtain the predicted positioning result, the predicted positioning result is compared with the test set, the error of the predicted positioning result is calculated, the model parameters are adjusted to reduce the error, and the performance of the prediction model is improved.

[0033] If the similarity between the collected environmental feature data and the historical environmental feature data is higher than a set threshold, the positioning result corresponding to the historical environmental feature data is matched with the current collected environmental feature data.

[0034] The environment-adaptive high-precision Beidou visual SLAM collaborative positioning optimization method also includes step 4, and real-time monitoring and adjustment of the positioning results, specifically: Real-time monitoring of positioning accuracy and environmental changes; According to the real-time monitoring results, the parameters of the environment adaptive model are dynamically adjusted to optimize the weight distribution of the visual SLAM and passive systems; The positioning error is corrected and data fusion is performed again according to the weight distribution to form a new unified positioning result.

[0035] In step 3, the visual SLAM data and the Beidou system data are fused according to their respective weight distributions through the Kalman filter, specifically: Step 3a, preprocessing the visual SLAM data and the Beidou system data so that the visual SLAM data and the Beidou system data have the same timestamp and coordinate system, and performing weighted processing on the visual SLAM data and the Beidou system data according to the weight distribution determined by the environment adaptive model; Step 3b, initializing the state vector and covariance matrix of the Kalman filter, the state vector includes position and velocity positioning parameters, and the covariance matrix is ​​used to represent the uncertainty of state estimation; Step 3c, using the time update equation of the Kalman filter, predicting the state estimate and covariance matrix at the current moment based on the state estimate and process noise covariance matrix at the previous moment; Step 3d, input the weighted visual SLAM data and Beidou system data into the Kalman filter as observation values, use the measurement update equation of the Kalman filter, and update the state estimate and covariance matrix according to the observation values ​​and the observation noise covariance matrix; Step 3e, repeat steps 3c to 3d until the required positioning accuracy or number of iterations is achieved, and output the final updated state estimate as a unified positioning result.

[0036] Specifically, an embodiment of Kalman filter fusing visual SLAM data and Beidou system data according to their respective weight distributions is as follows: initialization: The state vector X = [x, y, z, vx, vy, vz] includes three-dimensional position and three-dimensional velocity.

[0037] The covariance matrix P is initialized as a diagonal matrix, which represents the uncertainty of the state estimate.

[0038] Time update: Assuming that the process noise covariance matrix Q is a constant, the state estimate X^k|k-1 and covariance matrix P^k|k-1 at the current moment are predicted according to the time update equation of the Kalman filter.

[0039] Measurement Update: The weighted visual SLAM data and Beidou system data are input into the Kalman filter as the observation value Zk.

[0040] The observation noise covariance matrix R is dynamically adjusted according to the weight adjustment results of the environment adaptive model.

[0041] The state estimate X^k and the covariance matrix P^k are updated using the measurement update equation of the Kalman filter.

[0042] Iteration: The time update and measurement update steps are repeated until the desired positioning accuracy or number of iterations is achieved.

[0043] In order to verify the effectiveness of the present invention, the following experiments can be performed: Experiment on changing environmental conditions: Test the positioning accuracy of the system under different lighting conditions, feature point density and Beidou signal strength.

[0044] Experimental steps: Set up the experimental environment: select experimental scenes with representative different lighting conditions (such as strong light, weak light, drastic lighting changes, etc.), feature point density (such as high density, low density, medium density, etc.) and Beidou signal strength (such as strong signal, weak signal, signal obstruction, etc.).

[0045] Deploy sensors: Deploy sensor modules in the selected experimental scenario, including binocular cameras, Beidou satellite navigation receivers, INS (including IMU, etc.), etc., to ensure time synchronization between sensors.

[0046] Collect data: Start the sensor module to collect environmental feature data, visual SLAM data and Beidou system data.

[0047] Data processing and analysis: The environmental adaptation module is used to analyze the collected data, determine the environmental characteristics, and adjust the weights of the visual SLAM and Beidou system based on the analysis results. Then, the improved Kalman filter algorithm is used to fuse the data to obtain the positioning result.

[0048] Positioning accuracy evaluation: Compare the obtained positioning results with the actual position, calculate the positioning error, and evaluate the positioning accuracy of the system under different environmental conditions.

[0049] Experiments show that under conditions of strong light, high-density feature points and strong Beidou signals, the system has the highest positioning accuracy and the smallest error.

[0050] Under conditions of weak light, low-density feature points and weak Beidou signals, the system positioning accuracy decreases but remains within an acceptable range.

[0051] In complex environments such as drastic lighting changes, sparse feature points and Beidou signal obstruction, the system can still maintain a high positioning accuracy through environmental adaptive adjustment.

[0052] Dynamic environment experiment: Test the robustness and real-time adjustment effect of the system in a mobile environment.

[0053] Experimental steps: Set up the experimental scene: Select a moving scene that contains multiple environmental features (such as lighting changes, dynamic obstacles, etc.), such as city streets, parks, etc.

[0054] Deploy sensors: Deploy sensor modules in the selected experimental scenario and ensure time synchronization between sensors.

[0055] Start the system: Start the sensor module and real-time adjustment module, and start data collection and processing.

[0056] Dynamic monitoring and adjustment: During the movement, the positioning accuracy of the system and environmental changes are monitored in real time. According to the real-time monitoring results, the parameters of the environmental adaptive model are dynamically adjusted to optimize the weight distribution of the visual SLAM and Beidou system.

[0057] Error correction and evaluation: According to the results of real-time adjustment, the positioning error is corrected, and the robustness of the system and the real-time adjustment effect in a dynamic environment are evaluated.

[0058] Experiments show that the system can adjust the parameters of the environment adaptation model in real time and optimize the weight distribution during movement.

[0059] In complex environments such as lighting changes and dynamic obstacles, the system can maintain stable positioning accuracy and has strong robustness.

[0060] The real-time adjustment mechanism effectively improves the system's adaptability in dynamic environments and reduces positioning errors.

[0061] Comparative experiment: compare with the existing visual SLAM and Beidou system positioning methods to evaluate the performance improvement effect of the present invention.

[0062] Experimental steps: Selection of comparison method: Select the existing visual SLAM and Beidou system positioning methods as comparison objects.

[0063] Set up the experimental environment: Select representative experimental scenarios to ensure the consistency of the experimental environment.

[0064] Deploy sensors and algorithms: Deploy sensor modules in the selected experimental scenarios, and run the method of the present invention and the comparative method respectively.

[0065] Collect and process data: Collect environmental feature data, visual SLAM data and Beidou system data, and use two methods to process and locate them respectively.

[0066] Positioning accuracy evaluation: Compare the obtained positioning results with the actual position, calculate the positioning error, and evaluate the positioning accuracy of the two methods.

[0067] Experiments show that under the same environmental conditions, the positioning accuracy of the present invention is significantly higher than that of the existing visual SLAM and Beidou system positioning methods.

[0068] The present invention performs better in complex environments and significantly improves positioning accuracy.

[0069] Compared with the comparative method, the present invention is more robust and can adapt to more environmental changes.

[0070] The above-described embodiment is only a preferred solution of the present invention and does not limit the present invention in any form. There are other variations and modifications without exceeding the technical solution described in the claims.

Claims

1. An environment-adaptive high-precision Beidou visual SLAM collaborative positioning optimization method, characterized in that: The following steps are involved: Step 1, collect environmental feature data, visual SLAM data and Beidou system data; Step 2: construct an environment adaptive model, which adjusts the weight distribution of visual SLAM and Beidou system according to environmental feature data; Step 3: Fusion the visual SLAM data and BeiDou system data according to their respective weight distribution to form a unified positioning result.

2. A kind of environment-adaptive high-precision Beidou visual SLAM collaborative positioning optimization method according to claim 1, it is characterized in that, In the step 1, the collected environmental feature data include the current light intensity, the feature point distribution of the visual SLAM and the Beidou signal quality.

3. A kind of environment-adaptive high-precision Beidou visual SLAM collaborative positioning optimization method according to claim 2, it is characterized in that, The step 2 is specifically as follows: Constructing an environment adaptive model, the environment adaptive model analyzes the environment feature data to determine the characteristics of the current environment, wherein the characteristics of the current environment are: determining whether the current environment is well-lit or insufficiently lit according to the light intensity, judging the richness of the distribution of feature points in the current environment, and the strength and stability of the Beidou signal; According to the characteristics of the current environment, adjust the weight distribution of visual SLAM and Beidou system: in an environment with good lighting and rich distribution of feature points, increase the weight of visual SLAM; in an environment with insufficient lighting and insufficient distribution of feature points, reduce the weight of visual SLAM; in an environment with strong and stable Beidou signals, increase the weight of the Beidou system; in an environment with weak and unstable Beidou signals, reduce the weight of the Beidou system.

4. A kind of environment-adaptive high-precision Beidou visual SLAM collaborative positioning optimization method according to claim 3, characterized in that, The environmental adaptive model is also optimized to improve the adaptability and accuracy of the model through an adaptive learning mechanism. Specifically: Collect historical environmental characteristic data and pre-process the historical environmental characteristic data to remove noise and outliers; Build a prediction model and divide the historical environmental feature data into a training set and a test set. The training set is used to train the model, and the test set is used to evaluate the performance of the model. The training set is used to train the prediction model. Based on historical environmental feature data, the prediction model is used to obtain the predicted positioning result, the predicted positioning result is compared with the test set, the error of the predicted positioning result is calculated, the model parameters are adjusted to reduce the error, and the performance of the prediction model is improved.

5. A method for optimizing the collaborative positioning of high-precision Beidou visual SLAM based on an environment-adaptive method according to claim 4, characterized in that: If the similarity between the collected environmental feature data and the historical environmental feature data is higher than a set threshold, the positioning result corresponding to the historical environmental feature data is matched with the current collected environmental feature data.

6. A method for optimizing the collaborative positioning of high-precision Beidou visual SLAM based on an environment-adaptive method according to claim 1, characterized in that: The step 4 is also included, and the positioning result is monitored and adjusted in real time, specifically: Real-time monitoring of positioning accuracy and environmental changes; According to the real-time monitoring results, the parameters of the environment adaptive model are dynamically adjusted to optimize the weight distribution of the visual SLAM and passive systems; The positioning error is corrected and data fusion is performed again according to the weight distribution to form a new unified positioning result.

7. A method for optimizing the collaborative positioning of high-precision Beidou visual SLAM based on an environment-adaptive method according to claim 1, characterized in that: In step 3, the visual SLAM data and the Beidou system data are fused according to their respective weight distributions through the Kalman filter, specifically: Step 3a, preprocessing the visual SLAM data and the Beidou system data so that the visual SLAM data and the Beidou system data have the same timestamp and coordinate system, and performing weighted processing on the visual SLAM data and the Beidou system data according to the weight distribution determined by the environment adaptive model; Step 3b, initializing the state vector and covariance matrix of the Kalman filter, the state vector includes position and velocity positioning parameters, and the covariance matrix is ​​used to represent the uncertainty of state estimation; Step 3c, using the time update equation of the Kalman filter, predicting the state estimate and covariance matrix at the current moment based on the state estimate and process noise covariance matrix at the previous moment; Step 3d, input the weighted visual SLAM data and Beidou system data into the Kalman filter as observation values, use the measurement update equation of the Kalman filter, and update the state estimate and covariance matrix according to the observation values ​​and the observation noise covariance matrix; Step 3e, repeat steps 3c to 3d until the required positioning accuracy or number of iterations is achieved, and output the final updated state estimate as a unified positioning result.

Images