Autonomous positioning method based on visual inertial feature map multi-source fusion

Abstract

The invention belongs to the technical field of guidance. The invention provides a self-localization method based on visual inertial feature map multi-source fusion. Comprising the following steps: performing visual inertia fusion positioning on IMU information and visual image information to obtain a visual inertia fusion result; performing feature map matching positioning on the visual inertia fusion result to obtain a feature map matching result; and dynamically correcting the visual inertia fusion result and the feature map matching result to obtain a visual inertia feature map multi-source fusion positioning result. According to the embodiment of the invention, the feature map matching result and the estimation result of visual inertia fusion are dynamically corrected through error state Kalman filtering, and collaborative optimization of global and local information is realized. Performing effective correction on the drift of visual inertial positioning by using global constraints of the feature map; in a dynamic or unknown scene, continuity and real-time performance of positioning are maintained through visual inertia fusion, so that adaptability and stability of the system in a complex dynamic environment are enhanced.

Description

Technical Field

[0001] The disclosed embodiments relate to the field of guidance technology, and in particular to an autonomous positioning method based on multi-source fusion of visual inertial feature maps. Background Art

[0002] Simultaneous Localization and Mapping (SLAM) technology is the core technology for intelligent autonomous robots to perform mission objectives. In recent years, this technology has been widely studied and applied and has been successfully applied to many fields such as micro-UAVs, intelligent driving, virtual reality and augmented reality.

[0003] By capturing and processing rich environmental information, visual sensors can achieve feature extraction, map construction and high-precision positioning, significantly improving navigation performance. Visual SLAM does not rely on external navigation signals and can provide reliable positioning information even when GNSS signals are weak or unavailable, greatly enhancing the flexibility and safety of the task. However, visual SLAM is susceptible to factors such as environmental changes, unstable lighting, and sparse feature points during long-term operation, resulting in error accumulation. Especially in dynamic scenes, its positioning accuracy and stability are significantly challenged.

[0004] In contrast, the visual positioning method based on prior maps can efficiently and accurately achieve autonomous positioning by matching the real-time collected visual information with the pre-built feature map, and performs well in known scenes. However, in unknown or dynamically changing scenes, the pre-built map may cause positioning interruption due to insufficient information, thus affecting the stability and accuracy of the system.

[0005] Therefore, it is necessary to improve one or more problems existing in the above-mentioned related technical solutions.

[0006] It should be noted that this section is intended to provide background or context for the technical solutions of the present disclosure stated in the claims. The description herein is not admitted to be prior art by virtue of being included in this section. Summary of the invention

[0007] The purpose of the embodiments of the present disclosure is to provide an autonomous positioning method based on multi-source fusion of visual inertial feature maps, thereby overcoming one or more problems caused by the limitations and defects of related technologies at least to a certain extent.

[0008] According to an embodiment of the present disclosure, an autonomous positioning method based on multi-source fusion of visual inertial feature maps is provided, the method comprising: Perform visual-inertial fusion positioning on IMU information and visual image information to obtain visual-inertial fusion results; Perform feature map matching and positioning on the visual inertial fusion results to obtain feature map matching results; The visual-inertial fusion results and feature map matching results are dynamically corrected to obtain the multi-source fusion positioning result of the visual-inertial feature map.

[0009] Furthermore, the step of performing visual-inertial fusion positioning on the IMU information and the visual image information to obtain a visual-inertial fusion result includes: Pre-integrate the IMU information to obtain IMU pre-integral; Establishing IMU pre-integration residual according to IMU pre-integration; Performing image feature extraction on the visual image information of the current image frame to obtain image feature points of the current image frame; Match image feature points to establish visual reprojection residuals; A tightly coupled optimization model is established based on the IMU pre-integration residual and the visual reprojection residual; Based on the tightly coupled optimization model, the sliding window method is used to perform nonlinear optimization on the position and attitude of the carrier to obtain the visual-inertial fusion result.

[0010] Furthermore, the expression of the optimization variable in nonlinear optimization is:

[0011] in, is the IMU pre-integration residual, is the visual reprojection residual.

[0012] Furthermore, the optimization variables include the position, velocity, attitude, accelerometer bias, gyroscope bias and inverse depth of key points of all image frames within the sliding window.

[0013] Furthermore, the step of performing feature map matching and positioning on the visual inertial fusion result to obtain a feature map matching result includes: Calculate the Euclidean distance between the current image frame and each key map frame in the feature map library to obtain a coarse retrieval similarity measure; Calculate the Hamming distance between the current image frame and each key map frame in the feature map library to obtain a fine retrieval similarity measure; Using a coarse-to-fine hierarchical retrieval strategy, the current image frame is associated with each key map frame in the feature map library, and the feature map matching result is obtained through geometric constraints and optimization algorithms.

[0014] Furthermore, the expression of the rough retrieval similarity measure is:

[0015] in, The current image frame The global descriptor of Quantity, The feature map library The global descriptor of the keymap frame Quantity; The expression of fine retrieval similarity measurement is:

[0016] in, The current image frame No. i A local descriptor, The current image frame No. i The local descriptor t Quantity, The feature map library Keymap frame j A local descriptor, The feature map library Keymap frame j The local descriptor t components, and D is the total number of feature dimensions contained in each local descriptor.

[0017] Furthermore, the step of dynamically correcting the visual-inertial fusion result and the feature map matching result to obtain the visual-inertial feature map multi-source fusion positioning result includes: Design an error state Kalman filter and select the error state vector based on the visual inertial fusion results; The error state vector is propagated based on the motion model of visual-inertial fusion to obtain the state propagation equation; Based on the state propagation equation, the error state variable is predicted using the IMU integral value between two key map frames; A measurement model is constructed according to the feature map matching result, and the error state vector is updated according to the measurement model; The updated error state vector is used to dynamically correct the visual-inertial fusion result to obtain the multi-source fusion positioning result of the visual-inertial feature map.

[0018] The technical solution provided by the embodiments of the present disclosure may have the following beneficial effects: In the embodiments of the present disclosure, through the above-mentioned autonomous positioning method based on multi-source fusion of visual-inertial feature maps, on the one hand, a visual-inertial fusion tightly coupled optimization model is constructed, which can provide continuous and stable positioning solution results in dynamic environments, and significantly improve the robustness and adaptability of the system. Secondly, a coarse-to-fine hierarchical retrieval strategy is adopted to efficiently realize the rapid retrieval of the feature map library in large-scale scenes, and combined with 2D-2D epipolar geometry constraints, the global pose of the current image frame is accurately solved, effectively improving the accuracy and efficiency of positioning. On the other hand, an error state Kalman filter is designed to dynamically combine the feature map matching results with the estimation results of visual-inertial fusion, and realize the coordinated optimization of global and local information. In known scenes, the feature map provides a strong global constraint, which can effectively correct the accumulated drift of visual-inertial positioning; in dynamic or unknown scenes, visual-inertial fusion positioning ensures the continuity and real-time performance of the system, thereby improving the adaptability and stability of the system in complex dynamic environments. BRIEF DESCRIPTION OF THE DRAWINGS

[0019] The accompanying drawings herein are incorporated into the specification and constitute a part of the specification, illustrate embodiments consistent with the present disclosure, and together with the specification are used to explain the principles of the present disclosure. Obviously, the accompanying drawings described below are only some embodiments of the present disclosure, and for ordinary technicians in this field, other accompanying drawings can be obtained based on these accompanying drawings without creative work.

[0020] Figure 1 A step diagram of an autonomous positioning method based on multi-source fusion of visual-inertial feature maps in an exemplary embodiment of the present disclosure is shown; Figure 2 A specific flow chart of an autonomous positioning method based on multi-source fusion of visual inertial feature maps in an exemplary embodiment of the present disclosure is shown; Figure 3 A trajectory comparison diagram of a multi-source fusion positioning algorithm of a visual-inertial feature map in an exemplary embodiment of the present disclosure is shown. DETAILED DESCRIPTION

[0021] Example embodiments will now be described more fully with reference to the accompanying drawings. However, example embodiments can be implemented in a variety of forms and should not be construed as limited to the examples set forth herein; rather, these embodiments are provided so that the disclosure will be more comprehensive and complete and to fully convey the concepts of the example embodiments to those skilled in the art. The described features, structures, or characteristics may be combined in any suitable manner in one or more embodiments.

[0022] In addition, the accompanying drawings are only schematic illustrations of the embodiments of the present disclosure and are not necessarily drawn to scale. The same reference numerals in the figures represent the same or similar parts, and their repeated descriptions will be omitted. Some of the block diagrams shown in the accompanying drawings are functional entities and do not necessarily correspond to physically or logically independent entities.

[0023] This example implementation provides an autonomous positioning method based on multi-source fusion of visual inertial feature maps. Figure 1 As shown in , the autonomous positioning method based on multi-source fusion of visual inertial feature maps may include: steps S101 to S103.

[0024] Step S101: performing visual-inertial fusion positioning on the IMU information and the visual image information to obtain a visual-inertial fusion result; Step S102: performing feature map matching and positioning on the visual-inertial fusion result to obtain a feature map matching result; Step S103: dynamically correct the visual-inertial fusion result and the feature map matching result to obtain a visual-inertial feature map multi-source fusion positioning result.

[0025] Through the above-mentioned autonomous positioning method based on multi-source fusion of visual-inertial feature maps, on the one hand, a visual-inertial fusion tightly coupled optimization model is constructed, which can provide continuous and stable positioning solution results in dynamic environments, significantly improving the robustness and adaptability of the system. Secondly, a coarse-to-fine hierarchical retrieval strategy is adopted to efficiently realize the rapid retrieval of the feature map library in large-scale scenes, and combined with 2D-2D epipolar geometry constraints, the global pose of the current image frame is accurately solved, effectively improving the accuracy and efficiency of positioning. On the other hand, an error state Kalman filter is designed to dynamically combine the feature map matching results with the estimation results of visual-inertial fusion, realizing the coordinated optimization of global and local information. In known scenes, the feature map provides a strong global constraint, which can effectively correct the accumulated drift of visual-inertial positioning; in dynamic or unknown scenes, visual-inertial fusion positioning ensures the continuity and real-time performance of the system, thereby improving the adaptability and stability of the system in complex dynamic environments.

[0026] Next, we will refer to Figures 1 to 3 The various steps of the autonomous positioning method based on multi-source fusion of visual-inertial feature maps in this example implementation are described in more detail.

[0027] In step S101, visual-inertial fusion positioning is performed on the IMU information and the visual image information to obtain a visual-inertial fusion result.

[0028] Specifically, visual-inertial fusion positioning: First, the IMU (inertial measurement unit) information is pre-integrated, and feature points are extracted and matched from the visual image information. On this basis, the IMU pre-integration residual and the visual reprojection residual are calculated, and a tightly coupled optimization model is established. By constructing a sliding window, the position and posture of the carrier are optimized and solved. Within the sliding window, the optimized state quantities include the position, velocity, attitude, accelerometer bias, gyroscope bias, and inverse depth of all key points of all image frames. The specific state quantity definitions are as follows: ,in Indicates the number of image frames in the sliding window; Represents the total number of all keys in the sliding window; , where For the The image frame corresponds to the position of the carrier in the local coordinate system. For the Each image frame corresponds to the velocity of the carrier in the local coordinate system. For the The image frames correspond to the posture of the carrier in the local coordinate system. , Respectively represent the accelerometer bias and gyroscope bias of the IMU; Indicates The inverse depth of each key point.

[0029] In step S102, feature map matching and positioning are performed on the visual-inertial fusion result to obtain a feature map matching result.

[0030] Specifically, feature map matching and positioning: for each image frame, extract the global descriptor, and use the global features to quickly filter out the candidate key frames in the feature map library that are similar to the current frame. Then, calculate the Hamming distance of the local features between the current frame and the candidate key frames to complete the accurate retrieval of the feature map library. On this basis, solve the pose of the current image frame through 2D-2D epipolar geometry constraint optimization.

[0031] In step S103, the visual-inertial fusion result and the feature map matching result are dynamically corrected to obtain a visual-inertial feature map multi-source fusion positioning result.

[0032] Specifically, visual inertial feature map multi-source fusion positioning: design error state Kalman filter, select As the error state vector, where: is the position error vector, is the velocity error vector, is the attitude error vector, is the accelerometer bias error, is the gyroscope bias error, is the gravity acceleration error vector. First, the IMU data between image frames is used to predict the error state vector. Then, when the feature map matching positioning solution is successful, the matching result is used as the observation value to update the error state vector. Finally, the visual inertial fusion positioning result is used as the As the nominal state variable, the accumulated error is corrected in combination with the updated error state vector, thereby achieving the coordinated optimization of global and local information.

[0033] In a specific embodiment, Figure 2 The idea of ​​this application is demonstrated, including three main modules: visual-inertial fusion positioning, feature map matching positioning, and visual-inertial feature map multi-source fusion positioning. First, feature points of visual image information are extracted and matched, and IMU information is pre-integrated; then, the IMU pre-integration residual and visual reprojection residual are calculated, and a tightly coupled optimization model is established to obtain the visual-inertial fusion positioning result through optimization solution. At the same time, a coarse-to-fine hierarchical retrieval strategy is adopted to realize rapid retrieval of feature map libraries in large-scale scenes, and combined with 2D-2D epipolar geometry constraints, the global pose of the current image frame is accurately calculated. Finally, by designing an error state Kalman filter, the feature map matching results are dynamically combined with the estimation results of visual-inertial fusion to achieve collaborative optimization of global and local information.

[0034] Specific process: Visual inertial fusion positioning: First, the IMU information is pre-integrated, and feature points are extracted and matched from the visual image information. On this basis, the IMU pre-integration residual is calculated. and the visual reprojection residual , a tightly coupled optimization model is established, and the sliding window method is used to optimize the position and posture of the carrier. The optimization objective function is: .

[0035] The optimization variables include the position, velocity, attitude, accelerometer bias, gyroscope bias, and inverse depth of all key points of all image frames in the sliding window. The Levenberg-Marquardt algorithm is used to solve the optimization problem and obtain high-precision carrier pose estimation results. The specific optimization variables are defined as follows: ,in Indicates the number of image frames in the sliding window; Represents the total number of all keys in the sliding window; , where For the The image frame corresponds to the position of the carrier in the local coordinate system. For the Each image frame corresponds to the velocity of the carrier in the local coordinate system. For the The image frames correspond to the posture of the carrier in the local coordinate system. , Respectively represent the accelerometer bias and gyroscope bias of the IMU; Indicates The inverse depth of each key point.

[0036] Feature map matching positioning First, the Euclidean distance between the current image frame and each key map frame in the feature map is calculated as a similarity measure in the coarse retrieval stage:

[0037] in, The current image frame The global descriptor of Quantity, For the map library The global descriptor of the keymap frame A quantity.

[0038] By setting a similarity threshold To filter out the initial candidate keyframe set When the number of candidate keyframe sets Less than threshold , it means that there are not enough candidate frames in the map library and the current image retrieval fails; otherwise, we Sort the similarity scores of each map frame in the image and select the one with the highest similarity score. Frames as query image frames The candidate keyframe set ,in is a set of candidate key frames, They are the 1st, 2nd, 3rd, …, Mth candidate key frames respectively.

[0039] Secondly, the Hamming distance is calculated as the similarity measure in the detailed retrieval stage. The current image frame A set of descriptors, where is the number of descriptors; let For the A set of descriptors of candidate keymap frames, where is the number of descriptors. We calculate The sum of the Hamming distances between each descriptor in the candidate keymap frames and all descriptors in the current image frame :

[0040] in, is the current image frame The descriptor in and Descriptors in candidate keymap frames The Hamming distance between them is calculated as:

[0041] The smaller the Hamming distance, the more similar the features are. Calculate the sum of the Hamming distances of all descriptors between the current image frame and all candidate key frames in sequence. , select the key frame with the smallest value as the final matching frame.

[0042] Finally, the global pose of the current frame is solved by combining the 2D-2D epipolar geometry constraints.

[0043] Visual inertial feature map multi-source fusion positioning: First, the error state vector is selected as:

[0044] in, is the position error vector, is the velocity error vector, is the attitude error vector, is the accelerometer bias error, is the gyroscope bias error, is the gravity acceleration error vector.

[0045] Secondly, the error state vector is propagated based on the motion model of visual inertial fusion, and the state propagation equation of the system is obtained as follows: ,in, for k The state vector at time, for k-1 The state vector at time, is the nonlinear dynamic model of the state, is the IMU measurement value, is process noise, which obeys zero-mean Gaussian distribution and has a covariance of . Further linearization processing, we get ,in, is the state transfer matrix, which depends on the current state and control input, is the process noise distribution matrix.

[0046]

[0047] In the formula is the identity matrix, is the time interval between two frames of IMU data, for kThe rotation matrix at time, is the accelerometer measurement value, is the gyroscope measurement value, is an antisymmetric matrix.

[0048] The error state variable is predicted using the IMU integral value between two key frames:

[0049] in, and They are the mean square error matrix of state estimation at the previous moment and the mean square error matrix of state prediction at the current moment respectively.

[0050] Next, the feature map matching results are introduced to build the measurement model:

[0051] in, is the nonlinear function of the measurement, is the measurement noise, which obeys the zero-mean Gaussian distribution and has a covariance of Through linearization processing, the measurement error can be further expressed as:

[0052] in, is the measurement error, is the position obtained by visual inertial fusion positioning solution, The location obtained by matching and positioning the feature map, is the inverse of the attitude matrix obtained by visual inertial fusion positioning solution, is the posture matrix obtained by feature map matching and positioning. To convert the rotation matrix to an angle, measure the matrix for:

[0053] When the feature map matching and positioning is successful, the error state is updated:

[0054] in, is the Kalman gain, is the updated error state vector, is the mean square error matrix of the updated state estimate.

[0055] Finally, after the error state Kalman filter is completed, the visual inertial fusion positioning result is corrected in combination with the error state vector, and finally a high-precision fusion positioning result is obtained. The specific implementation of error correction is as follows:

[0056] in, is the final fusion positioning result. is the current system status, is the attitude error quaternion, expressed as: .

[0057] Through the above-mentioned implementation steps, the present application realizes efficient and stable multi-source fusion positioning, significantly improving the positioning accuracy and system stability in complex dynamic environments.

[0058] Figure 3 The experimental running trajectory comparison between the visual-inertial feature map multi-source fusion positioning algorithm and the visual-inertial fusion positioning algorithm in a complex dynamic environment is shown. In the figure, the solid line "— RTK" represents the positioning trajectory obtained by the high-precision real-time kinematic measurement (RealTime Kinematic, RTK) system, which serves as the benchmark for accuracy comparison; the dotted line "···GLNS" represents the trajectory of the visual-inertial feature map multi-source fusion positioning algorithm; "—·— VIO" is the trajectory of the visual-inertial fusion positioning algorithm. The experimental results show that the visual-inertial fusion positioning algorithm accumulates large errors and produces obvious trajectory drift during long-term operation, while the trajectory of the visual-inertial feature map multi-source fusion positioning algorithm proposed in this application is almost completely consistent with the benchmark trajectory RTK, proving that the algorithm can achieve continuous, stable and accurate navigation positioning in a complex dynamic environment.

[0059] Through the above-mentioned autonomous positioning method based on multi-source fusion of visual-inertial feature maps, on the one hand, a visual-inertial fusion tightly coupled optimization model is constructed, which can provide continuous and stable positioning solution results in dynamic environments, significantly improving the robustness and adaptability of the system. Secondly, a coarse-to-fine hierarchical retrieval strategy is adopted to efficiently realize the rapid retrieval of the feature map library in large-scale scenes, and combined with 2D-2D epipolar geometry constraints, the global pose of the current image frame is accurately solved, effectively improving the accuracy and efficiency of positioning. On the other hand, an error state Kalman filter is designed to dynamically combine the feature map matching results with the estimation results of visual-inertial fusion, realizing the coordinated optimization of global and local information. In known scenes, the feature map provides a strong global constraint, which can effectively correct the accumulated drift of visual-inertial positioning; in dynamic or unknown scenes, visual-inertial fusion positioning ensures the continuity and real-time performance of the system, thereby improving the adaptability and stability of the system in complex dynamic environments.

[0060] It should be understood that the terms "center", "longitudinal", "lateral", "length", "width", "thickness", "up", "down", "front", "back", "left", "right", "vertical", "horizontal", "top", "bottom", "inside", "outside", "clockwise", "counterclockwise", etc. in the above description indicate orientations or positional relationships based on the orientations or positional relationships shown in the accompanying drawings, and are only for the convenience of describing the embodiments of the present disclosure and simplifying the description, and do not indicate or imply that the referred device or element must have a specific orientation, be constructed and operated in a specific orientation, and therefore should not be understood as a limitation on the embodiments of the present disclosure.

[0061] In addition, the terms "first" and "second" are used for descriptive purposes only and should not be understood as indicating or implying relative importance or implicitly indicating the number of technical features indicated. Thus, a feature defined as "first" or "second" may explicitly or implicitly include one or more of the features. In the description of the embodiments of the present disclosure, the meaning of "multiple" is two or more, unless otherwise clearly and specifically defined.

[0062] In the embodiments of the present disclosure, unless otherwise clearly specified and limited, the terms "installed", "connected", "connected", "fixed" and the like should be understood in a broad sense, for example, it can be a fixed connection, a detachable connection, or an integral connection; it can be a mechanical connection or an electrical connection; it can be a direct connection or an indirect connection through an intermediate medium, it can be the internal connection of two elements or the interaction relationship between two elements. For ordinary technicians in this field, the specific meanings of the above terms in the present disclosure can be understood according to specific circumstances.

[0063] In the embodiments of the present disclosure, unless otherwise clearly specified and limited, a first feature being "above" or "below" a second feature may include the first and second features being in direct contact, or may include the first and second features not being in direct contact but being in contact through another feature between them. Moreover, a first feature being "above", "above" and "above" a second feature includes the first feature being directly above and obliquely above the second feature, or simply indicates that the first feature is higher in level than the second feature. A first feature being "below", "below" and "below" a second feature includes the first feature being directly below and obliquely below the second feature, or simply indicates that the first feature is lower in level than the second feature.

[0064] In the description of this specification, the description with reference to the terms "one embodiment", "some embodiments", "example", "specific example" or "some examples" etc. means that the specific features, structures, materials or characteristics described in conjunction with the embodiment or example are included in at least one embodiment or example of the present disclosure. In this specification, the schematic representations of the above terms do not necessarily refer to the same embodiment or example. Moreover, the specific features, structures, materials or characteristics described may be combined in any one or more embodiments or examples in a suitable manner. In addition, those skilled in the art may combine and combine the different embodiments or examples described in this specification.

[0065] Those skilled in the art will readily appreciate other embodiments of the present disclosure after considering the specification and practicing the invention disclosed herein. This application is intended to cover any modification, use or adaptation of the present disclosure, which follows the general principles of the present disclosure and includes common knowledge or customary techniques in the art that are not disclosed in the present disclosure. The specification and examples are intended to be exemplary only, and the true scope and spirit of the present disclosure are indicated by the appended claims.

Claims

1. An autonomous positioning method based on multi-source fusion of visual inertial feature maps, characterized in that: The method includes: Perform visual-inertial fusion positioning on IMU information and visual image information to obtain visual-inertial fusion results; Perform feature map matching and positioning on the visual inertial fusion results to obtain feature map matching results; The visual-inertial fusion results and feature map matching results are dynamically corrected to obtain the multi-source fusion positioning result of the visual-inertial feature map.

2. The autonomous positioning method based on multi-source fusion of visual inertial feature maps according to claim 1 is characterized in that: The steps of performing visual-inertial fusion positioning on the IMU information and the visual image information to obtain the visual-inertial fusion result include: Pre-integrate the IMU information to obtain IMU pre-integral; Establishing IMU pre-integration residual according to IMU pre-integration; Performing image feature extraction on the visual image information of the current image frame to obtain image feature points of the current image frame; Match image feature points to establish visual reprojection residuals; A tightly coupled optimization model is established based on the IMU pre-integration residual and the visual reprojection residual; Based on the tightly coupled optimization model, the sliding window method is used to perform nonlinear optimization on the position and attitude of the carrier to obtain the visual-inertial fusion result.

3. The autonomous positioning method based on multi-source fusion of visual inertial feature maps according to claim 2 is characterized in that: The expression of the optimization variable in nonlinear optimization is: in, is the IMU pre-integration residual, is the visual reprojection residual.

4. The autonomous positioning method based on multi-source fusion of visual inertial feature maps according to claim 3 is characterized in that: The optimization variables include the position, velocity, attitude, accelerometer bias, gyroscope bias, and inverse depth of key points for all image frames within the sliding window.

5. The autonomous positioning method based on multi-source fusion of visual inertial feature maps according to claim 4 is characterized in that: The step of performing feature map matching and positioning on the visual inertial fusion result to obtain the feature map matching result includes: Calculate the Euclidean distance between the current image frame and each key map frame in the feature map library to obtain a coarse retrieval similarity measure; Calculate the Hamming distance between the current image frame and each key map frame in the feature map library to obtain a fine retrieval similarity measure; Using a coarse-to-fine hierarchical retrieval strategy, the current image frame is associated with each key map frame in the feature map library, and the feature map matching result is obtained through geometric constraints and optimization algorithms.

6. The autonomous positioning method based on multi-source fusion of visual inertial feature maps according to claim 5 is characterized in that: The expression of rough retrieval similarity measurement is: in, The current image frame The global descriptor of Quantity, The feature map library The global descriptor of the keymap frame Quantity; The expression of fine retrieval similarity measurement is: in, The current image frame No. i A local descriptor, The current image frame No. i The local descriptor t Quantity, The feature map library Keymap frame j A local descriptor, The feature map library Keymap frame j The local descriptor t components, and D is the total number of feature dimensions contained in each local descriptor.

7. The autonomous positioning method based on multi-source fusion of visual inertial feature maps according to claim 6 is characterized in that: The step of dynamically correcting the visual-inertial fusion result and the feature map matching result to obtain the visual-inertial feature map multi-source fusion positioning result includes: Design an error state Kalman filter and select the error state vector based on the visual inertial fusion results; The error state vector is propagated based on the motion model of visual-inertial fusion to obtain the state propagation equation; Based on the state propagation equation, the error state variable is predicted using the IMU integral value between two key map frames; A measurement model is constructed according to the feature map matching result, and the error state vector is updated according to the measurement model; The updated error state vector is used to dynamically correct the visual-inertial fusion result to obtain the multi-source fusion positioning result of the visual-inertial feature map.

Images