A geomagnetic fingerprint map dynamic maintenance method based on a mobile robot

Abstract

This invention relates to the field of indoor geomagnetic positioning technology, specifically to a method for dynamic maintenance of geomagnetic fingerprint maps based on mobile robots. The method includes multimodal data acquisition and spatiotemporal alignment to establish a correspondence between sensor data and global map coordinates; constructing a cross-confidence mechanism using geomagnetic anomaly factors and visual verification factors, and accurately triggering map updates using a time decay accumulation model; identifying newly added magnetic objects and extracting their parameters, calculating the radius of influence of the disturbance field using a magnetic field disturbance model; and employing a region-growing local incremental update method, introducing a boundary smoothing factor to ensure map continuity. This invention achieves autonomous and efficient maintenance of geomagnetic fingerprint maps without requiring global rescanning or additional hardware, improving the stability of geomagnetic positioning for mobile robots. Simultaneously, the accurate triggering of updates through a dual-modal cross-confidence evaluation mechanism enhances the accuracy of geomagnetic fingerprint map updates.

Description

Technical Field

[0001] This invention relates to the field of indoor geomagnetic positioning technology, specifically to a method for dynamic maintenance of geomagnetic fingerprint maps based on a mobile robot. Background Technology

[0002] Geomagnetic positioning technology relies on a pre-constructed geomagnetic fingerprint map. By matching real-time collected geomagnetic data with the geomagnetic fingerprint map, the location of the target can be calculated. Geomagnetic signals have advantages such as high stability, no obstruction, and no cumulative error. They are widely used in the positioning and navigation of indoor mobile robots. Their positioning accuracy directly depends on the consistency between the fingerprint map and the real-time environmental magnetic field distribution.

[0003] However, in real-world indoor applications, the addition, removal, or relocation of ferromagnetic materials can significantly alter the local magnetic field distribution characteristics. This environmental disturbance causes pre-generated geomagnetic fingerprints to quickly become invalid, leading to matching errors, positioning drift, and other problems. Current technologies require periodic re-collection of global data to reconstruct the fingerprint, resulting in poor real-time performance and failing to meet the online maintenance needs in dynamic environments. Detection of environmental changes relies on single magnetometer data and fixed threshold judgments, lacking verification of the physical properties of the change source and making it difficult to distinguish between long-term changes and transient disturbances. Therefore, developing an efficient geomagnetic fingerprint maintenance mechanism that accurately identifies long-term magnetic disturbances is a pressing issue that needs to be addressed. Summary of the Invention

[0004] To address the technical problems existing in the prior art, this invention provides a mobile robot-based method for dynamic maintenance of geomagnetic fingerprint maps, which achieves autonomous and efficient maintenance without requiring global rescanning or additional hardware, improves the stability of geomagnetic positioning of mobile robots, and enhances the accuracy of geomagnetic fingerprint map updates by accurately triggering updates through a dual-modal cross-confidence evaluation mechanism.

[0005] To achieve the above objectives, the technical solution adopted in this invention is a dynamic maintenance method for geomagnetic fingerprint maps based on a mobile robot, including a magnetometer and a vision sensor mounted on a mobile robot, and constructing an original geomagnetic fingerprint map of the target indoor environment, which is operated according to the following steps.

[0006] S1. By moving a mobile robot in the target indoor environment, the three-dimensional raw magnetic field data of the magnetometer, the color and depth images of the RGB-D vision sensor, as well as the timestamp and pose information are collected synchronously at a fixed frequency. The precise correspondence between the sensor measurement data and the global map coordinates is established to form a multimodal data frame associated with the map coordinates.

[0007] S2. Construct a piecewise linear geomagnetic anomaly factor using the Pearson correlation coefficient; analyze color and depth images using a visual target detection algorithm to extract the detection confidence of newly added ferromagnetic objects as a visual verification factor; multiply the geomagnetic anomaly factor and the visual verification factor to obtain the single observation confidence; accumulate the confidence of multiple observations using a time decay model to obtain the spatiotemporal cumulative confidence of the mobile robot's location; when the cumulative confidence exceeds a preset update threshold, determine that the mobile robot's location has undergone a continuous environmental change, mark it as a seed point, and trigger the map update process.

[0008] S3. Based on the visual and depth images in the multimodal data frame, identify the type, geometric parameters and spatial coordinates of the newly added ferromagnetic object. According to the object type, call the preset magnetic field disturbance model to calculate the magnetic field disturbance value generated by the newly added object at each point in space. Based on the environmental magnetic field noise level, search the values ​​to determine the influence radius of the magnetic field disturbance. With the seed point as the center and the influence radius as the boundary, delineate the effective range of map update.

[0009] S4. Regional growth-type local incremental update introduces a boundary smoothing factor to achieve a smooth transition between the updated and unupdated regions. It traverses each grid point within the region to be updated, superimposes the original geomagnetic fingerprint value with the magnetic field disturbance value weighted by the boundary smoothing factor, and obtains the updated geomagnetic fingerprint value. The grid points outside the updated region retain their original fingerprint values, thus completing the local incremental update of the geomagnetic fingerprint map.

[0010] Preferably, in step S1, the original magnetic field data of the magnetometer, the color image and depth image of the RGB-D vision sensor are acquired simultaneously, and the acquired data are preprocessed and spatiotemporally aligned in combination with the pose information of the wheel odometer.

[0011] Preferably, the preprocessing includes calibration and total field strength calculation. Spatiotemporal alignment refers to aligning magnetometer measurement data, visual image data, and pose information in time and space, corresponding them to the same grid coordinates on the global map, forming a set of spatially unique multi-source data, with each grid point corresponding to a set of data.

[0012] The calibration formula is as follows: ,

[0013] In the formula, The three-dimensional raw magnetic field data collected by the magnetometer This is the offset of the hard iron. The error matrix of soft iron The inverse matrix, The calibrated three-dimensional magnetic field data;

[0014] After calibration, the total magnetic field strength is calculated using the following formula: ,

[0015] In the formula, The x-axis components of the calibrated data. The y-axis components of the calibrated data. The z-axis components of the calibrated data. This represents the total magnetic field strength after calibration.

[0016] After preprocessing and spatiotemporal alignment, the data forms a multimodal data frame associated with global map coordinates, including: map coordinates. Total magnetic field strength after calibration Color images and depth images.

[0017] Preferably, in step S2, based on the multimodal data frames, the geomagnetic anomaly factor and visual verification factor are calculated, the total magnetic field strength sequence of k consecutive frames at the current location coordinates is extracted, and the Pearson correlation coefficient r between this sequence and the reference sequence at the corresponding location of the original geomagnetic fingerprint is calculated. The calculation formula is as follows: ,

[0018] In the formula, r is the Pearson correlation coefficient. This represents the i-th value of the total field strength sequence for the current k frames. The mean of the current k-frame sequence. For the i-th value of the reference sequence, The mean of the baseline sequence;

[0019] Piecewise linear geomagnetic anomaly factors were constructed based on the Pearson correlation coefficient r. As shown in the following formula:

[0020] ,

[0021] The color images acquired by the RGB-D vision sensor are then identified using a visual verification factor, and the detection confidence score for newly added magnetic or ferromagnetic objects is output. This detection confidence score is the visual verification factor. ;

[0022] The confidence level of a single observation is obtained by fusing geomagnetic verification factors and visual verification factors. The spatiotemporal cumulative confidence score is obtained by combining the time decay model, as shown in the following formula:

[0023] , ,

[0024] In the formula, It is a geomagnetic anomaly factor. As a visual verification factor, Let be the spatiotemporal cumulative confidence score of the current frame t. The initial value is the spatiotemporal cumulative confidence score of the previous frame. , The time decay coefficient, The confidence level of a single observation in the current frame t is used. When the cumulative confidence level exceeds the preset threshold k, it is marked as a seed point.

[0025] Preferably, in step S3, based on the multimodal data frame, the type, geometric parameters and spatial coordinates of the newly added ferromagnetic object are identified, a preset magnetic field disturbance model is called, the magnetic field disturbance value generated by the object at each point in space is calculated, and the area to be updated on the map is delineated.

[0026] Preferably, the preset magnetic field disturbance model includes a magnetic dipole model or a linear magnetic charge model. When a new ferromagnetic object is a cube or a sphere, the magnetic dipole model is used, and the magnetic field disturbance value is... The calculation formula is: ,

[0027] In the formula, The permeability of free space, Let be the magnetic moment of the object. Let be the vector from a point in space to the center of the object. For vectors The length of the mold, For vectors and The dot product;

[0028] The elongated object utilizes a linear magnetic charge model to determine the magnetic field disturbance value. The calculation formula is:

[0029] ,

[0030] In the formula, The permeability of free space, This is the equivalent linear magnetic charge density. This represents the perpendicular distance from a point in space to the near end of the object. The perpendicular distance from a point in space to the far end of the object;

[0031] Then, a magnetic field noise threshold is preset based on the ambient background noise level. Search The maximum distance, The maximum distance is the magnetic field disturbance value of the object, which is the radius of influence R of the magnetic field disturbance. With the marked seed point as the center and R as the radius, a circular area is drawn on the global map. This circular area is the area to be updated.

[0032] Preferably, in step S4, based on the defined region to be updated, and with the marked seed point as the center, the boundary smoothing factor of each grid point within the region is calculated using the following formula: ,

[0033] In the formula, For boundary smoothing factor, The straight-line distance from the grid point to the seed point within the area to be updated. The radius of influence of the magnetic field disturbance;

[0034] Then adjust the boundary smoothing factor and magnetic field disturbance value In combination, the geomagnetic fingerprint values ​​of each grid point within the area are locally updated using the following formula: ,

[0035] In the formula, The updated geomagnetic fingerprint value, This is the original geomagnetic fingerprint value. This represents the magnetic field disturbance value for that grid point.

[0036] Compared with the prior art, the present invention has the following technical effects: The present invention accurately triggers updates through a dual-modal cross-confidence evaluation mechanism, avoiding the problem of false detection by a single sensor; it calls a preset magnetic field disturbance model based on the recognition result of newly added magnetic objects to calculate the magnetic field disturbance range; and it adopts a region-growing local update method to achieve efficient updating of the fingerprint map, enabling the mobile robot to autonomously detect newly added magnetic objects and locally update the geomagnetic fingerprint map in a dynamic environment, maintaining the accuracy of the fingerprint map without global rescanning. Attached Figure Description

[0037] Figure 1 This is a flowchart of the present invention.

[0038] Figure 2 This is a schematic diagram of the local update of the geomagnetic fingerprint in this invention. Detailed Implementation

[0039] To make the technical problems to be solved, the technical solutions, and the beneficial effects of the present invention clearer, the present invention will be further described in detail below with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative of the present invention and are not intended to limit the present invention.

[0040] like Figure 1 As shown, a method for dynamic maintenance of geomagnetic fingerprint maps based on mobile robots includes a magnetometer and a vision sensor mounted on a mobile robot, as well as the construction of an original geomagnetic fingerprint map of the target indoor environment. The method is operated according to the following steps:

[0041] S1. Multimodal data acquisition and preprocessing: During operation, the mobile robot synchronously acquires three-dimensional raw magnetic field data from the magnetometer, color images and depth images from the RGB-D vision sensor at a fixed frequency, calibrates the magnetic field data to obtain the real-time total magnetic field strength, and obtains real-time pose information through the robot's own wheel odometry. Based on the timestamp and pose information, all data are spatiotemporally aligned to establish a precise correspondence between sensor measurement data and global map coordinates, forming a multimodal data frame associated with map coordinates.

[0042] S2. Cross-confidence assessment and update triggering: Extract the magnetic field time series at the current map coordinates, calculate its Pearson correlation coefficient with the historical baseline sequence of the corresponding location in the original geomagnetic fingerprint map, and construct a piecewise linear geomagnetic anomaly factor based on the correlation coefficient; analyze color and depth images through visual target detection algorithms, and extract the detection confidence of newly added ferromagnetic objects as a visual verification factor; multiply the geomagnetic anomaly factor and the visual verification factor to obtain the single observation confidence, and accumulate the confidence of multiple observations using a time decay model to obtain the spatiotemporal cumulative confidence of the mobile robot's location; when the cumulative confidence exceeds the preset update threshold, it is determined that the mobile robot's location has undergone continuous environmental change, marked as a seed point, and the map update process is triggered.

[0043] S3. Change source identification and magnetic field disturbance modeling: Based on visual and depth images in multimodal data frames, identify the type, geometric parameters and spatial coordinates of newly added ferromagnetic objects. According to the object type, call the preset magnetic field disturbance model to calculate the magnetic field disturbance value generated by the newly added object at each point in space. Based on the environmental magnetic field noise level, search values ​​to determine the influence radius of the magnetic field disturbance. With the seed point as the center and the influence radius as the boundary, delineate the effective range of map update.

[0044] S4. Regional growth-type local incremental update introduces a boundary smoothing factor to achieve a smooth transition between the updated and unupdated regions. It traverses each grid point within the region to be updated, superimposes the original geomagnetic fingerprint value with the magnetic field disturbance value weighted by the boundary smoothing factor, and obtains the updated geomagnetic fingerprint value. The grid points outside the updated region retain their original fingerprint values, thus completing the local incremental update of the geomagnetic fingerprint map.

[0045] In specific operations,

[0046] I. Multimodal Data Acquisition and Preprocessing

[0047] During its movement within a known indoor environment, the mobile robot simultaneously acquires raw magnetic field data from a magnetometer, color images from an RGB-D vision sensor, and depth images. Combined with pose information from its own wheeled odometer, the acquired data undergoes preprocessing and spatiotemporal alignment. Preprocessing includes calibration and total field strength calculation. Spatiotemporal alignment refers to aligning the magnetometer measurement data, visual image data, and pose information in time and space, mapping them to the same grid coordinates on the global map. In this process, a set of multi-source data with unique spatial locations is formed, with each grid point corresponding to a set of data.

[0048] 1. Magnetic field data calibration: To address practical issues such as sensor vibration and interference from metal parts of the robot during operation, magnetic field calibration is completed through hard iron offset and soft iron error compensation to eliminate system errors.

[0049] The calibration formula is as follows: ,

[0050] In the formula, The three-dimensional raw magnetic field data collected by the magnetometer This is the offset of the hard iron. The error matrix of soft iron The inverse matrix, This is the calibrated three-dimensional magnetic field data.

[0051] 2. Total Field Strength Calculation: The calibrated three-dimensional data is converted into the total magnetic field strength, which serves as the core indicator for geomagnetic feature matching.

[0052] After calibration, the total magnetic field strength is calculated using the following formula: ,

[0053] In the formula, The x-axis components of the calibrated data. The y-axis components of the calibrated data. The z-axis components of the calibrated data. This represents the total field strength of the calibrated magnetic field.

[0054] 3. Spatiotemporal Alignment: Based on the data acquisition timestamp and wheel odometer pose information, magnetic field data and visual image data are accurately mapped to the same grid coordinates on the global map. This forms a multimodal data frame with associated map coordinates, including: map coordinates Total magnetic field strength after calibration Color images and depth images. In practical applications, the raster resolution can be adjusted according to the scene's positioning accuracy requirements to ensure accurate correspondence between data and physical space.

[0055] II. Cross-confidence assessment and update triggering (addressing the high false positive rate issue)

[0056] By fusing geomagnetic anomaly factors and visual verification factors in a dual-modal manner, combined with a time decay accumulation model, we can accurately determine the continuous magnetic disturbances in the environment, and completely solve the practical problems of false detection by a single sensor and invalid updates triggered by instantaneous interference.

[0057] 1. Geomagnetic Anomaly Factor Calculation: Extract the total magnetic field strength sequence of k consecutive frames at the current location coordinates, calculate the Pearson correlation coefficient r between this sequence and the reference sequence at the corresponding location in the original geomagnetic fingerprint, and quantify the similarity between the real-time magnetic field and the reference magnetic field. The calculation formula is as follows:

[0058] ,

[0059] In the formula, r is the Pearson correlation coefficient. This represents the i-th value of the total field strength sequence for the current k frames. The mean of the current k-frame sequence. For the i-th value of the reference sequence, The mean of the baseline sequence;

[0060] Piecewise linear geomagnetic anomaly factors were constructed based on the Pearson correlation coefficient r. Regions with low magnetic field similarity are weighted and amplified to improve anomaly detection sensitivity, as shown in the following formula:

[0061] ,

[0062] 2. Visual Validation Factor Acquisition: This is achieved through a pre-trained object detection algorithm model, which is already deployed in the system. Lightweight YOLOv5 / YOLOv8 models can be used to adapt to the robot's edge computing capabilities. The model identifies newly added ferromagnetic objects in the image, and then uses the visual validation factor to identify the color image acquired by the RGB-D vision sensor, outputting the detection confidence score of the newly added magnetic ferromagnetic object. This detection confidence score is the visual validation factor. The higher the value, the higher the reliability of the detection of newly added ferromagnetic objects.

[0063] 3. Single observation confidence fusion: The geomagnetic anomaly factor is multiplied with the visual verification factor to achieve dual-modal cross-validation and obtain the single observation confidence. As shown in the following formula: In the formula, It is a geomagnetic anomaly factor. As a visual verification factor,

[0064] The confidence level of a single observation will only increase significantly when the geomagnetic data detects an anomaly and the visual data detects a new ferromagnetic object, thus avoiding misjudgment caused by a single magnetometer due to transient magnetic field interference (such as a metal object passing by temporarily).

[0065] 4. Spatiotemporal cumulative confidence calculation: A time decay model is introduced to accumulate the confidence scores of multiple observations, resulting in a spatiotemporal cumulative confidence score. This score adaptively distinguishes between continuous environmental changes and temporary disturbances. The calculation formula is as follows:

[0066] ,

[0067] In the formula, Let be the spatiotemporal cumulative confidence score of the current frame t. The initial value is the spatiotemporal cumulative confidence score of the previous frame. , The time decay coefficient, Let be the confidence level of a single observation in the current frame t. The larger the value, the higher the weight of historical observations, and the better it can suppress transient interference.

[0068] 5. Update trigger judgment: When the spatiotemporal cumulative confidence of a certain grid point is updated. When the preset update threshold k is exceeded, it is determined that a continuous magnetic disturbance has occurred at that location, and it is marked as a seed point, triggering the map update process.

[0069] III. Identification of sources of change and modeling of magnetic field disturbances (precisely defining the update range)

[0070] The physical properties of magnetic disturbance sources are identified based on multimodal data frames. The disturbance value and influence radius are calculated by combining the preset magnetic field disturbance model. The area to be updated is accurately delineated, avoiding invalid global updates and reducing the robot's computing resource consumption.

[0071] 1. Magnetic disturbance source identification: Accurately identify the type, geometric parameters, and spatial coordinates of newly added ferromagnetic objects using color and depth images from an RGB-D vision sensor; distinguish between cubes / spheres and elongated shapes to provide a basis for selecting magnetic field disturbance models; obtain the actual size and three-dimensional spatial coordinates of objects through depth images to provide accurate parameters for model calculation.

[0072] 2. Magnetic field disturbance value calculation: Based on the object type, a preset magnetic field disturbance model is invoked to calculate the magnetic field disturbance value generated by the object at each grid point in space. It adapts to the magnetic field distribution characteristics of ferromagnetic objects of different shapes. The preset magnetic field perturbation models include magnetic dipole models or linear magnetic charge models.

[0073] When a new ferromagnetic object is added as a cube or sphere, the magnetic dipole model is invoked to match the magnetic field radiation characteristics of the bulk metal object, and the magnetic field perturbation value is adjusted. The calculation formula is: ,

[0074] In the formula, The permeability of free space, Let be the magnetic moment of the object. Let be the vector from a point in space to the center of the object. For vectors The length of the mold, For vectors and The dot product;

[0075] When the newly added ferromagnetic object is a long, thin object, the linear magnetic charge model is invoked to match the magnetic field radiation characteristics of the long, thin metal object, and the magnetic field perturbation value is adjusted. The calculation formula is:

[0076] ,

[0077] In the formula, The permeability of free space, This is the equivalent linear magnetic charge density. This represents the perpendicular distance from a point in space to the near end of the object. It represents the perpendicular distance from a point in space to the far end of the object.

[0078] 3. Determining the radius of influence of disturbance: A magnetic field noise threshold is preset based on the ambient background noise level. Search The maximum distance, The maximum distance is the magnetic field disturbance value of the object, which is the radius of influence R of the magnetic field disturbance. With the marked seed point as the center and R as the radius, a circular area is drawn on the global map. This circular area is the area to be updated.

[0079] IV. Regional Growth-Based Local Incremental Updates (Ensuring Location Continuity)

[0080] By introducing a boundary smoothing factor, the defined area to be updated is locally incrementally updated to achieve a smooth transition between the updated and unupdated areas, avoiding robot positioning drift and jumps caused by sudden changes in the fingerprint map, and ensuring positioning stability.

[0081] 1. Boundary Smoothing Factor Calculation: Based on the defined region to be updated, and centered on the marked seed point, calculate the boundary smoothing factor for each grid point within the region. The calculation formula is as follows: ,

[0082] In the formula, For boundary smoothing factor, The straight-line distance from the grid point to the seed point within the area to be updated. The radius of influence of the magnetic field disturbance is denoted as . The closer the grid point is to the seed point, the larger the weighting coefficient of the magnetic field disturbance value; the closer the grid point is to the boundary of the update region, the smaller the weighting coefficient of the magnetic field disturbance value, thus achieving gradient decay of the update amplitude.

[0083] 2. Local fingerprint update: Traverse each grid point in the area to be updated, and superimpose the original geomagnetic fingerprint value with the magnetic field disturbance value after being weighted by the boundary smoothing factor to obtain the updated geomagnetic fingerprint value. The grid points outside the update area keep the original fingerprint value unchanged, thus completing the local incremental update of the geomagnetic fingerprint map.

[0084] Boundary smoothing factor and magnetic field disturbance value In combination, the geomagnetic fingerprint values ​​of each grid point within the area are locally updated using the following formula: ,

[0085] In the formula, The updated geomagnetic fingerprint value, This is the original geomagnetic fingerprint value. This represents the magnetic field disturbance value for that grid point.

[0086] The updated fingerprint map achieves gradient changes in magnetic field values ​​in the boundary region. When the robot moves to the boundary between the updated and unupdated regions, the positioning results do not change significantly, ensuring the smoothness of the operation process.

[0087] This invention proposes a cross-confidence mechanism that fuses geomagnetic anomaly factors and visual verification factors. By mutually verifying geomagnetic and visual information, it achieves accurate and reliable judgment of environmental changes. Furthermore, it constructs a time-decaying cumulative confidence model to adaptively distinguish between continuous environmental changes and temporary disturbances, improving the reliability of update triggering. Simultaneously, it proposes a regional growth-based local repair update method based on a magnetic field disturbance model, updating only the areas affected by magnetic field disturbances to avoid global reconstruction. Finally, it introduces a distance-based linear smoothing factor to control the update amplitude, ensuring the spatial continuity of local map updates and preventing boundary jumps.

[0088] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions and improvements made within the spirit and principles of the present invention should be included within the scope of the present invention.

Claims

1. A method for dynamic maintenance of geomagnetic fingerprint maps based on mobile robots, comprising a magnetometer and a vision sensor mounted on a mobile robot, and constructing an original geomagnetic fingerprint map of a target indoor environment, characterized in that: Follow these steps: S1. By moving a mobile robot in the target indoor environment, the three-dimensional raw magnetic field data of the magnetometer, the color and depth images of the RGB-D vision sensor, as well as the timestamp and pose information are collected synchronously at a fixed frequency. The precise correspondence between the sensor measurement data and the global map coordinates is established to form a multimodal data frame associated with the map coordinates. S2. Construct piecewise linear geomagnetic anomaly factors using Pearson correlation coefficient; analyze color and depth images using visual target detection algorithms to extract the detection confidence of newly added ferromagnetic objects as visual verification factors. The confidence level of a single observation is obtained by multiplying the geomagnetic anomaly factor and the visual verification factor. The confidence levels of multiple observations are accumulated by combining the time decay model to obtain the spatiotemporal cumulative confidence level of the location of the mobile robot. When the cumulative confidence level exceeds the preset update threshold, it is determined that the location of the mobile robot has undergone a continuous environmental change, which is marked as a seed point and the map update process is triggered. S3. Based on the visual and depth images in the multimodal data frame, identify the type, geometric parameters and spatial coordinates of the newly added ferromagnetic object, call the preset magnetic field disturbance model according to the object type, and calculate the magnetic field disturbance value generated by the newly added object at each point in space. Based on the environmental magnetic field noise level, the search values ​​determine the radius of influence of the magnetic field disturbance, and the effective range of map update is defined with the seed point as the center and the radius of influence as the boundary. S4. Regional growth-type local incremental update introduces a boundary smoothing factor to achieve a smooth transition between the updated and unupdated regions. It traverses each grid point within the region to be updated, superimposes the original geomagnetic fingerprint value with the magnetic field disturbance value weighted by the boundary smoothing factor, and obtains the updated geomagnetic fingerprint value. The grid points outside the updated region retain their original fingerprint values, thus completing the local incremental update of the geomagnetic fingerprint map.

2. The method for dynamic maintenance of geomagnetic fingerprint maps based on mobile robots according to claim 1, characterized in that: In step S1, the original magnetic field data of the magnetometer, the color image and depth image of the RGB-D vision sensor are collected simultaneously, and the collected data are preprocessed and spatiotemporally aligned in combination with the pose information of the wheel odometer.

3. The method for dynamic maintenance of geomagnetic fingerprint maps based on mobile robots according to claim 2, characterized in that: The preprocessing includes calibration and total field strength calculation. Spatiotemporal alignment refers to aligning magnetometer measurement data, visual image data, and pose information in time and space, corresponding them to the same grid coordinates on the global map, forming a set of spatially unique multi-source data, with each grid point corresponding to a set of data. The calibration formula is as follows: , In the formula, The three-dimensional raw magnetic field data collected by the magnetometer This is the offset of the hard iron. The error matrix of soft iron The inverse matrix, The calibrated three-dimensional magnetic field data; After calibration, the total magnetic field strength is calculated using the following formula: , In the formula, The x-axis components of the calibrated data. The y-axis components of the calibrated data. The z-axis components of the calibrated data. This represents the total magnetic field strength after calibration. After preprocessing and spatiotemporal alignment, the data forms a multimodal data frame associated with global map coordinates, including: map coordinates. Total magnetic field strength after calibration Color images and depth images.

4. The method for dynamic maintenance of geomagnetic fingerprint maps based on mobile robots according to claim 3, characterized in that: In step S2, based on the multimodal data frames, the geomagnetic anomaly factor and visual verification factor are calculated, the total magnetic field strength sequence of k consecutive frames at the current location coordinates is extracted, and the Pearson correlation coefficient r between this sequence and the reference sequence at the corresponding location of the original geomagnetic fingerprint is calculated. The calculation formula is as follows: , In the formula, r is the Pearson correlation coefficient. This represents the i-th value of the total field strength sequence for the current k frames. The mean of the current k-frame sequence. The i-th value of the reference sequence The mean of the baseline sequence; Piecewise linear geomagnetic anomaly factors were constructed based on the Pearson correlation coefficient r. As shown in the following formula: ， The color images acquired by the RGB-D vision sensor are then identified using a visual verification factor, and the detection confidence score for newly added magnetic or ferromagnetic objects is output. This detection confidence score is the visual verification factor. ; The confidence level of a single observation is obtained by fusing geomagnetic verification factors and visual verification factors. The spatiotemporal cumulative confidence score is obtained by combining the time decay model, as shown in the following formula: ， ， In the formula, It is a geomagnetic anomaly factor. As a visual verification factor, Let be the spatiotemporal cumulative confidence score of the current frame t. The initial value is the spatiotemporal cumulative confidence score of the previous frame. , The time decay coefficient, The confidence level of a single observation in the current frame t is used. When the cumulative confidence level exceeds the preset threshold k, it is marked as a seed point.

5. The method for dynamic maintenance of geomagnetic fingerprint maps based on mobile robots according to claim 3, characterized in that: In step S3, based on the multimodal data frame, the type, geometric parameters and spatial coordinates of the newly added ferromagnetic object are identified, a preset magnetic field disturbance model is called, the magnetic field disturbance value generated by the object at each point in space is calculated, and the area to be updated on the map is delineated.

6. The method for dynamic maintenance of geomagnetic fingerprint maps based on mobile robots according to claim 5, characterized in that: The preset magnetic field disturbance model includes a magnetic dipole model or a linear magnetic charge model. When a new ferromagnetic object is a cube or a sphere, the magnetic dipole model is used, and the magnetic field disturbance value is... The calculation formula is: , In the formula, The permeability of free space, The magnetic moment of the object, Let be the vector from a point in space to the center of the object. For vectors The length of the mold, For vectors and The dot product; The elongated object utilizes a linear magnetic charge model to determine the magnetic field disturbance value. The calculation formula is: ， In the formula, The permeability of free space, This is the equivalent linear magnetic charge density. This represents the perpendicular distance from a point in space to the near end of the object. This is the perpendicular distance from a point in space to the far end of the object; Then, a magnetic field noise threshold is preset based on the ambient background noise level. Search The maximum distance, The maximum distance is the magnetic field disturbance value of the object, which is the radius of influence R of the magnetic field disturbance. With the marked seed point as the center and R as the radius, a circular area is drawn on the global map. This circular area is the area to be updated.

7. The method for dynamic maintenance of geomagnetic fingerprint maps based on mobile robots according to claim 6, characterized in that: In step S4, based on the defined region to be updated, and with the marked seed point as the center, the boundary smoothing factor of each grid point within the region is calculated using the following formula: , In the formula, For boundary smoothing factor, The straight-line distance from the grid point to the seed point within the area to be updated. The radius of influence of the magnetic field disturbance; Then adjust the boundary smoothing factor and magnetic field disturbance value In combination, the geomagnetic fingerprint values ​​of each grid point within the area are locally updated using the following formula: , In the formula, The updated geomagnetic fingerprint value, This is the original geomagnetic fingerprint value. This represents the magnetic field disturbance value for that grid point.

Images