Stereo vision inertial odometer positioning method and system based on quantization window

Abstract

The application provides a stereo vision inertial odometer positioning method and system based on a quantization window, and relates to the field of computer vision processing. The stereo vision inertial odometer positioning method based on the quantization window comprises a quantization window concept for abstracting straight cruising of an agricultural robot along a crop row; the latest driving state of the robot is quantized at a key frame level, and a corresponding driving state quantization index is obtained; according to the quantization index, a current quantization window is continuously separated and constructed from a historical trajectory of the robot; in a current quantization window resetting stage, a reference quantization window is selectively constructed based on the constructed current quantization window; a spatial parallel constraint between the quantization windows is constructed; spatial parallel constraint abnormalities between the quantization windows are detected, and a corresponding correction mechanism is proposed. The method solves the problem that a visual odometer cannot establish long-term data association in an agricultural field scene, and effectively reduces the global positioning cumulative error of the system.

Description

Technical Field

[0001] This invention relates to the field of computer vision processing technology, specifically to a stereo vision inertial odometry positioning method and system based on a quantization window. Background Technology

[0002] In the field of autonomous navigation for agricultural robots, a large amount of past research has opted for high-precision real-time dynamic global navigation satellite systems (RTK-GNSS) to obtain the precise location of agricultural robots. However, the high cost of the sensors themselves and the need for supporting ground base stations significantly raise the barrier to practical application of this method. Signal obstruction caused by crops or weather conditions in unstructured environments also affects the reliability of such solutions in real-world applications. Furthermore, the estimation of the robot's own attitude and the perception of its surrounding environment are additional issues that satellite positioning solutions need to consider.

[0003] Simultaneous Localization and Mapping (SLAM) is a core task in the perception phase of autonomous robot navigation. It estimates the robot's current pose and constructs a point cloud map of the environment using information from internal and external sensors. In recent years, many SLAM systems have expanded from classic indoor and urban scenarios to the agricultural field. Among them, visual SLAM has received widespread research due to its rich environmental information input and low cost. However, due to the complexity of the agricultural environment, accurate navigation of vision-based agricultural robots remains a challenging problem.

[0004] In agricultural settings, researchers have made adaptive improvements to common visual SLAM frameworks. These improvements can be broadly categorized into crop-dependent and crop-independent types. The utilization of crop features has led to the development of visual SLAM / visual odometry systems specifically designed for particular planting scenarios. These methods extract reliable semantic features of plants from visual images and perform short-term and long-term data association based on these features to achieve accurate localization and mapping. However, such methods often lack feasibility in open field scenarios, especially in low-lying crop fields with highly homogenized environments.

[0005] In non-crop-dependent improvement systems, researchers often focus on enhancing the front-end performance of SLAM systems in agricultural scenarios, addressing issues such as depth suppression, tracking failure and recovery, and feature selection. While these improvements can enhance system robustness and improve localization and mapping accuracy, they still struggle to provide long-term data associations when loop closure detection systems fail in agricultural settings. Furthermore, researchers also attempt to achieve reliable and accurate localization using artificial landmarks; however, deploying dense artificial landmarks over large areas in agricultural fields is not feasible.

[0006] In summary, suppressing the cumulative error of global positioning of odometry in agricultural field scenarios remains an unsolved problem. Summary of the Invention

[0007] (a) Technical problems to be solved To address the shortcomings of existing technologies, this invention provides a stereo vision inertial odometry positioning method and system based on a quantization window, solving the problems of visual odometry's inability to establish long-term data associations and the need to reduce global positioning cumulative errors in agricultural field scenarios.

[0008] (II) Technical Solution To achieve the above objectives, the present invention provides the following technical solution: Firstly, a stereo vision-based inertial odometry positioning method based on a quantization window is provided, including: A quantitative window abstraction concept for abstracting agricultural robots that cruise along crop rows; Acquire the latest keyframes obtained from the visual inertial odometry through preliminary calculation and estimation; The latest keyframe is input into the preset quantization window subsystem to quantize the robot's driving state at the keyframe level and obtain the corresponding driving state quantization index. Based on the driving status quantification index, the current quantification window is continuously separated and constructed from the robot's historical trajectory; During the current quantization window reset phase, a reference quantization window is selectively constructed based on the existing current quantization window. Construct spatial parallel constraints between the current quantization window and the reference quantization window; Detect anomalies in spatial parallel constraint matching relationships, correct abnormal sub-windows in spatial parallel constraint matching relationships, and finally complete map updates.

[0009] Preferably, the step of inputting the latest keyframe into a preset quantization window subsystem to quantize the robot's latest driving state at the keyframe level and obtain the corresponding driving state quantization index specifically includes: In the pre-defined quantization window subsystem, maintain a window of length... The latest keyframe queue was used, and the original loop closure detection task was masked; simulations were performed based on the structural characteristics of crop rows in the field. The theoretical location of the nearest keyframe, where, The length of the temporary window in the current quantization window; the reverse recursive formula for the theoretical position of each most recent keyframe in the most recent keyframe queue is as follows: in, and These represent the theoretical position and the actual estimated position of the latest keyframe, respectively. Represents the first two dimensions of the matrix. Indicates the horizontal distance between two locations; The robot's driving state at the moment of each latest keyframe is quantified by calculating the horizontal trajectory error between the actual keyframe sequence and the simulated keyframe sequence; the formula for calculating the Quantitative Index of Driving State (QIDS) is as follows: in, and These represent the actual estimated pose of the system and the simulated theoretical pose, respectively. This represents the projection of the translation vector of the pose matrix onto the XY plane; It represents the square of the Euclidean norm of a vector.

[0010] Preferably, the quantization window separation specifically includes: A threshold α is set to distinguish significant changes in the robot's driving state. If the driving state quantization index is less than α, the robot is considered to be maintaining a straight cruising state along the crop path, and the latest keyframe is added to the current quantization window. If the driving state quantization index is greater than or equal to α, the current frame is marked, and the most recent temporary window is analyzed in the current quantization window. If the keyframes marked in the temporary window meet the requirement of temporal continuity, it is determined that the robot's driving state has changed significantly, and the current quantization window will be reset.

[0011] Preferably, the construction of spatial parallel constraints between the current quantization window and the reference quantization window specifically includes: Parallel constraint matching is performed on each added keyframe in the current quantization window. The matching must meet the following conditions: the keyframe with the smallest optical center distance from the current keyframe in the reference quantization window is selected as the matching frame; the matching frame is located at a non-edge of the reference quantization window; the first keyframe in the current quantization window to complete spatial parallel constraint matching will be used as the starting keyframe for subsequent anomaly detection and correction methods. The system is allowed to have multiple reference quantization windows, which do not overlap vertically. It is also stipulated that the current quantization window can only establish frame-to-frame parallel constraint relationships with one reference quantization window. Specifically, when a significant change in the robot's driving state is detected, the reset of the current quantization window will be delayed due to the time continuity determination; therefore, several keyframes newly added to the current quantization window will be preferentially pruned. When the current quantization window is about to be reset, if there are consecutive sub-windows in the current quantization window that have not completed parallel constraint matching, and the length of the sub-windows reaches a certain threshold... If a keyframe in the current quantization window matches a different reference quantization window than the previous one, the window separation of the current quantization window ends, and a new current quantization window is created immediately.

[0012] Preferably, the detection of anomalies in the spatial parallel constraint matching relationship specifically includes: When the length of the sub-windows with parallel constraint matching relationships in the current quantization window is sufficient, the anomaly detection mechanism will be activated. The parallel constraint discrimination formula is as follows: in, and These represent the horizontal distances from the starting keyframe and the current keyframe to the optical center of their matching keyframe, respectively; λ is the threshold controlling the discrimination sensitivity; when the above inequality does not hold, the system will further verify the orientation angle of the quantization window to prevent false positives. in, This represents the unit vector indicating the overall direction of the current quantization window. The unit vector represents the overall direction of the matched reference quantization window; the frequency and arrangement of the high driving state quantization index in the temporary window must be much lower than the requirement of time continuity in order to ultimately meet the judgment of spatial parallel relationship anomalies.

[0013] Preferably, the correction of abnormal sub-windows in the spatial parallel constraint matching relationship specifically includes: The system obtains the relative position transformation vector ∆t between keyframe pairs based on the starting keyframe in the current quantization window and the matching keyframe of that frame in the reference quantization window, and then infers the corrected position vector of the current keyframe that triggered this correction in the current quantization window. : in, This represents the position vector of the current keyframe matched within the reference quantization window; since the quantization window correction algorithm focuses on eliminating the cumulative error of robot yaw angle estimation, the algorithm uses the original keyframe position vector... The component of gravity in the direction of gravity is retained, and the final vector is derived. .

[0014] By referring to the coordinate system of the starting keyframe in the current quantization window, we obtain and Unit vector in the principal coordinate system: In the formula: Represents the principal coordinate system; in formula (7), using and Solve for the axis angles required to correct the principal coordinate system. : pass Further, the pose transformation matrix required for correction is obtained. : In formula (8), the exponential mapping is expressed as ,in It is the rotation matrix required for quantization window correction; using The poses of all keyframes in sub-windows with parallel constraint matching relationships within the current quantization window are corrected and propagated; among them, The pose transformation of each corrected keyframe relative to the world coordinate system The following can be calculated: The transformation relationship between the velocity vectors of each keyframe and the principal coordinate system remains unchanged before and after correction. Obtain the corrected velocity vectors for each keyframe in the world coordinate system. : After the keyframes are corrected, the position of the point cloud is corrected according to the transformation relationship between the keyframes and the map point cloud.

[0015] Secondly, a stereo vision inertial odometry positioning system based on a quantization window is provided, the system comprising: The receiving module acquires the latest keyframes obtained by the visual inertial odometry through preliminary calculation and estimation; The quantization module is used to input the latest keyframe into the preset quantization window subsystem, quantify the latest driving state of the agricultural robot at the keyframe level, and obtain the corresponding driving state quantization index. The separation module continuously separates and constructs the current quantization window from the robot's historical trajectory based on the driving state quantization index, and selectively constructs a reference quantization window based on the constructed current quantization window during the current quantization window reset phase. The parallel constraint module is used to construct spatial parallel constraints between the current quantization window and the reference quantization window. The detection and correction module is used to detect anomalies in spatial parallel constraint matching relationships, correct the sub-windows of anomalies in spatial parallel constraint matching relationships, and complete map updates.

[0016] Thirdly, a computer-readable storage medium is provided for storing one or more programs, the one or more programs including instructions that, when executed by a computing device, cause the computing device to perform any of the methods described.

[0017] Fourthly, a computing device is provided, comprising: One or more processors, a memory, and one or more programs, wherein the one or more programs are stored in the memory and configured to be executed by the one or more processors, and the one or more programs include instructions for performing any of the methods described.

[0018] (III) Beneficial Effects This invention presents a stereo vision-based inertial odometry localization method and system based on quantized windows. It successfully designs an effective alternative to closed-loop detection algorithms in agricultural field scenarios and effectively reduces the cumulative error of the system's global localization. The method first proposes the concept of a quantized window to describe the robot's driving state as it cruises along crop rows. A driving state quantification algorithm is designed to quantify the agricultural robot's driving state, and then the quantized windows between crop rows are separated. The system constructs spatial parallel constraints based on the geometric relationships between the quantized windows and continuously maintains these geometric relationships through an anomaly correction algorithm. Attached Figure Description

[0019] Figure 1 This is a schematic diagram of the stereo vision inertial odometry positioning method based on a quantization window according to the present invention. Figure 2 This is a schematic diagram illustrating the construction of spatial parallel constraints in an embodiment of the present invention; Figure 3 This is a schematic diagram of the driving state quantization algorithm in an embodiment of the present invention; Figure 4 This is a flowchart of the quantization window separation algorithm in an embodiment of the present invention; Figure 5 This is a schematic diagram of the quantization window correction algorithm in an embodiment of the present invention. Detailed Implementation

[0020] The technical solutions in the embodiments of the present invention will now be clearly and completely described with reference to the accompanying drawings. Obviously, the described embodiments are only a part of the embodiments of the present invention, and not all of them. All other embodiments obtained by those skilled in the art based on the embodiments of the present invention without creative effort are within the scope of protection of the present invention.

[0021] Implementation, for example Figure 1As shown, this embodiment of the invention provides a stereo vision inertial odometry (VIO) localization method based on quantization windows. In the VIO system, the vision front-end retains the initialization strategy of ORB-SLAM3, achieving robust tracking through binocular images and IMU input. In the loop closure and map fusion thread of the ORB-SLAM3 system, the original loop closure detection task is masked. As a replacement and improvement, a quantization window subsystem is introduced into this thread to receive the latest keyframes provided by the local mapping thread. Throughout the tracking process, the system continuously maintains the spatial parallelism between quantization windows and performs map and trajectory correction updates at necessary times. The method specifically includes: The latest keyframe from the local mapping thread is received in the system's loopback and map fusion thread; The latest keyframe is input into the preset quantization window subsystem to quantize the robot's latest driving state at the keyframe level, and obtain the corresponding driving state quantization index; that is, the driving state quantization algorithm. Based on the driving status quantization index, the current quantization window is continuously separated and constructed from the robot's historical trajectory; during the current quantization window reset phase, a reference quantization window is selectively constructed based on the constructed current quantization window. Construct spatial parallel constraints between the current quantization window and the reference quantization window; Detect anomalies in spatial parallel constraint matching relationships, correct abnormal sub-windows in spatial parallel constraint matching relationships, and finally complete map updates.

[0022] Furthermore, the driving state quantization algorithm specifically involves embedding the proposed quantization window subsystem into the loop closure and map fusion thread of the ORB-SLAM3 system, thus masking the original loop closure detection task (which fails in this application scenario). Within the pre-defined quantization window subsystem, a window of length [length missing] is maintained. The most recent keyframe queue. The most recent keyframes were simulated based on the structural characteristics of crop rows in the field. The theoretical positions of each keyframe. Formula (1) represents the reverse recursion of the theoretical positions of each keyframe in the queue: in and These represent the theoretical and estimated positions of the most recent keyframe, respectively. Represents the first two dimensions of the matrix. This represents the horizontal distance between two locations. Under the initial conditions of location recursion, the theoretically estimated locations and the actual estimated locations of the two most recent keyframes are exactly the same, such as... Figure 3 .

[0023] exist Figure 3The diagram illustrates the difference between the actual trajectory (solid line) and the theoretical trajectory (dashed line) under different driving states (U-turn and straight cruising). The motion state of the agricultural robot is quantified by calculating the horizontal trajectory error between the actual and simulated keyframe sequences. The formula for calculating the driving state quantification index is as follows: in, and These represent the actual estimated pose of the system and the simulated theoretical pose, respectively. This represents the projection of the translation vector of the pose matrix onto the XY plane. This represents the square of the Euclidean norm of the vector. Furthermore, the positional information of keyframes in the gravity direction is not considered for two main reasons: 1) The driving state quantization algorithm primarily aims to describe the changes in the horizontal heading of the agricultural robot as accurately as possible; 2) In agriculture, due to the robot's bumpy movement, the gravity component often becomes the main source of error in the entire trajectory. This can lead to frequent false positives in the quantization algorithm, thus affecting the reliability of the quantization window separation.

[0024] Furthermore, the system considers a portion of the robot's trajectory along the crop rows as a quantization window, and selects a portion of the window that meets the requirements as a reference quantization window. To avoid false positives (unlike the false positives in the quantization algorithm above), spatial parallel constraints are constructed between windows.

[0025] The quantization window separation process is as follows Figure 4 As shown, the completion of the first inertial initialization of the system is taken as the starting point of this process. In the construction loop of the quantization window, the newly arrived keyframe is first obtained from the local mapping thread, and the driving quantization index (QIDS) for each keyframe is calculated. A threshold α is set to distinguish significant changes in the robot's driving state. If the QIDS is less than α, it is considered that the robot is maintaining a straight cruising state along the crop line and moving towards... Add the latest keyframe. Otherwise, it is considered that the driving state of the current frame may have changed significantly. To prevent false positives, the algorithm does not immediately give a judgment of driving state change, but marks the current frame and adds the latest keyframe. Analyze the most recent temporary window (e.g.) Figure 2 (As shown). If the keyframes marked in this temporary window meet the requirement of temporal continuity, then it is determined that the robot's driving state has changed significantly. It will be reset.

[0026] Furthermore, regarding Parallel constraint matching is performed on each added keyframe. The matching must meet the following two conditions: 1) Select The keyframe with the smallest optical center distance to the current keyframe is selected as the matching frame. 2) The matching frame should not be located in... The edge. Additionally. The first keyframe that completes spatial parallel constraint matching will serve as the starting keyframe for subsequent anomaly detection and correction methods.

[0027] To construct parallel constraints for quantitative windows in a large-scale agricultural scenario, a multi-reference window strategy was adopted in the system. The following stipulations were made: 1) The system can have multiple... And they do not overlap in the vertical direction of the window; 2) Only one is allowed Establish parallel constraints between frames. This is because the temporal continuity determination in the quantization window separation algorithm leads to… The reset delay means that newly added items will be pruned first. Several keyframes. When the system is about to reset. At that time, if There are consecutive sub-windows in the data where parallel constraint matching has not been completed, and the length of the sub-windows reaches [a certain value]. Then set the child window as In order to comply with the above provision 2), if The keyframe in the match is different from the past. ,but The window detachment ends, and a new one is created immediately. The effect of constructing spatial parallel constraints is as follows: Figure 2 As shown.

[0028] Furthermore, the anomaly correction mechanism will detect anomalies in the current spatial parallel matching relationships. When When the length of the sub-windows with parallel constraint matching relationships is sufficient, the anomaly detection mechanism will be activated. The parallel constraint discrimination formula is as follows: in, and These represent the horizontal distances from the starting keyframe and the current keyframe to the optical center of their matching keyframe, respectively. λ is a threshold controlling the discrimination sensitivity. When the above inequality does not hold, the system will further verify the orientation angle of the quantization window to prevent false positives. in, express The unit vector in the overall direction. Indicates the matched Furthermore, it is necessary to ensure that the frequency of high quantization exponents (QIDS) appearing in the temporary window is much lower than the requirement for temporal continuity (as mentioned in the quantization window separation algorithm). Because... The reset delay during window separation may cause false positives in the anomaly detection algorithm (the second type of false positive in this paper). The quantization window correction algorithm will be performed only if all the above conditions are met.

[0029] Furthermore, the quantization window correction process is as follows: Figure 5 As shown. The purpose of the correction algorithm is to... Correcting sub-windows with abnormal spatial parallelism in order to maintain... and The spatial parallel relationship between them. In (5), according to The starting keyframe and the frame in The relative position transformation vector between the matched keyframes Infer the corrected position vector of the current keyframe. ,Right now The corrected position vector of the keyframe that triggered this correction. Indicates the current keyframe is The position vectors of the keyframes matched in the algorithm. Since the quantization window correction algorithm focuses on eliminating the cumulative error of the robot's yaw angle estimation, the algorithm uses the original keyframe position vectors... The component of gravity in the direction of gravity is retained, and the final vector is derived. : It should be noted that the position vectors mentioned above are all referenced to the world coordinate system. The subsequent algorithm steps, however, refer to... The coordinate system (principal coordinate system) of the starting keyframe. This allows us to obtain... and Unit vector in the principal coordinate system: In the formula: Indicates the principal coordinate system. In (7), using... and Solve for the axis angles required to correct the principal coordinate system. .

[0030] pass Further, the pose transformation matrix required for correction is obtained. : In formula (8), the exponential mapping is expressed as ,in This is the rotation matrix required for quantization window correction. Then, using... right The poses of all keyframes in the window are corrected and propagated. Since the relative poses between keyframes within the window do not change before and after correction in the principal coordinate system, i.e. The pose transformation of each corrected keyframe relative to the world coordinate system. The following can be calculated: Similar to pose correction, the transformation relationship between the velocity vectors of each keyframe and the principal coordinate system remains unchanged before and after correction. This allows us to obtain the corrected velocity vectors of each keyframe in the world coordinate system. : After the keyframes are corrected, the point cloud position needs to be corrected based on the transformation relationship between the keyframes and the map point cloud.

[0031] in, The current quantization window; For reference quantization window; for The length of the temporary window in the process; To and Relevant length thresholds; This is the current quantization index for the robot's driving state quantization algorithm.

[0032] Embodiments of this application may be provided as methods or computer program products. Therefore, this application may take the form of a completely hardware embodiment, a completely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, this application may take the form of a computer program product implemented on one or more computer-usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code. The solutions in the embodiments of this application may be implemented in various computer languages, such as the object-oriented programming language Java and the interpreted scripting language JavaScript.

[0033] This application is described with reference to flowchart illustrations and / or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of this application. It will be understood that each block of the flowchart illustrations and / or block diagrams, and combinations of blocks in the flowchart illustrations and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, special-purpose computer, embedded processor, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, generate instructions for implementing the flowchart... Figure 1 One or more processes and / or boxes Figure 1 A device that provides the functions specified in one or more boxes.

[0034] These computer program instructions may also be stored in a computer-readable storage medium that can direct a computer or other programmable data processing device to function in a particular manner, such that the instructions stored in the computer-readable storage medium produce an article of manufacture including instruction means, which are implemented in a process Figure 1 One or more processes and / or boxes Figure 1 The function specified in one or more boxes.

[0035] These computer program instructions may also be loaded onto a computer or other programmable data processing equipment to cause a series of operational steps to be performed on the computer or other programmable equipment to produce a computer-implemented process, thereby providing instructions that execute on the computer or other programmable equipment for implementing the process. Figure 1 One or more processes and / or boxes Figure 1 The steps of the function specified in one or more boxes.

[0036] It should be noted that, in this document, relational terms such as "first" and "second" are used only to distinguish one entity or operation from another, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Furthermore, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, method, article, or apparatus. Without further limitations, an element defined by the phrase "comprising one..." does not exclude the presence of other identical elements in the process, method, article, or apparatus that includes said element.

Claims

1. A stereo vision inertial odometry positioning method based on a quantization window, characterized in that, include: Acquire the latest keyframes obtained from the visual inertial odometry through preliminary calculation and estimation; The latest keyframe is input into the preset quantization window subsystem to quantize the robot's latest driving state at the keyframe level and obtain the corresponding driving state quantization index. Based on the driving status quantification index, the current quantification window is continuously separated and constructed from the robot's historical trajectory; During the current quantization window reset phase, a reference quantization window is selectively constructed based on the existing current quantization window. Construct spatial parallel constraints between the current quantization window and the reference quantization window; Detect anomalies in spatial parallel constraint matching relationships, correct the anomaly sub-windows in spatial parallel constraint matching relationships, and complete map updates; The step of inputting the latest keyframe into a preset quantization window subsystem to quantize the robot's latest driving state at the keyframe level and obtain the corresponding driving state quantization index specifically includes: In the pre-defined quantization window subsystem, maintain a window of length... The most recent keyframe queue; simulation based on the structural characteristics of crop rows in the field. The theoretical location of the nearest keyframe, where, This represents the length of the temporary window in the current quantization window; the reverse recursive formula for the theoretical position of each most recent keyframe in the most recent keyframe queue is as follows: in, and These represent the theoretical and estimated positions of the most recent keyframe, respectively. Represents the first two dimensions of the matrix. Indicates the horizontal distance between two locations; The robot's driving state at the moment of the latest keyframe is quantified by calculating the horizontal trajectory error between the actual keyframe sequence and the simulated keyframe sequence; the formula for calculating the driving state quantization index is as follows: in, and These represent the actual estimated pose of the system and the simulated theoretical pose, respectively. This represents the projection of the translation vector of the pose matrix onto the XY plane; It represents the square of the Euclidean norm of a vector.

2. The stereo vision inertial odometry positioning method based on a quantization window according to claim 1, characterized in that: Quantization window separation, specifically including: A threshold α is set to distinguish whether the robot's driving state has changed significantly. If the driving state quantization index is less than α, it is considered that the robot is maintaining a straight cruise along the crop line, and the latest keyframe is added to the current quantization window. If the driving state quantization index is greater than or equal to α, the current frame is marked, and the most recent temporary window is analyzed in the current quantization window. If the keyframe marked in the temporary window meets the requirement of temporal continuity, it is determined that the robot's driving state has changed significantly, and the current quantization window will be reset.

3. The stereo vision inertial odometry positioning method based on a quantization window according to claim 1, characterized in that: The construction of spatial parallel constraints between the current quantization window and the reference quantization window specifically includes: Parallel constraint matching is performed on each keyframe added in the current quantization window. The matching must meet the following conditions: the keyframe with the smallest optical center distance from the current keyframe in the reference quantization window is selected as the matching frame; the matching frame is located at a non-edge of the reference quantization window; the first keyframe in the current quantization window to complete spatial parallel constraint matching will be used as the starting keyframe for subsequent anomaly detection and correction methods. The system specifies that it has multiple reference quantization windows that do not overlap vertically; it also specifies that the current quantization window can only establish frame-to-frame parallel constraint relationships with one reference quantization window; it prioritizes pruning keyframes newly added to the current quantization window; when the current quantization window is about to be reset, if there are consecutive sub-windows in the current quantization window that have not completed parallel constraint matching, and the length of the sub-windows reaches... Then, set the sub-window as the reference quantization window, where, This represents the length threshold associated with the reference quantization window; if a keyframe in the current quantization window matches a reference quantization window that is different from the past, the separation and construction of the current quantization window ends, and a new current quantization window is created immediately.

4. The stereo vision inertial odometry positioning method based on a quantization window according to claim 1, characterized in that: The anomalies in the detection space parallel constraint matching relationship specifically include: If the sub-windows with parallel constraint matching relationships in the current quantization window are long enough, the anomaly detection mechanism will be activated. The parallel constraint discrimination formula is as follows: in, and These represent the horizontal distances from the starting keyframe and the current keyframe to the optical center of their matching keyframe, respectively; λ is the threshold controlling the discrimination sensitivity; when the above inequality does not hold, the system will further verify the orientation angle of the quantization window to prevent false positives. in, This represents the unit vector indicating the overall direction of the current quantization window. This represents the unit vector of the overall direction of the matched reference quantization window; in addition, the frequency and arrangement of the high driving state quantization index in the temporary window must be much lower than the requirement of time continuity in order to ultimately meet the judgment of spatial parallel relationship anomalies.

5. The stereo vision inertial odometry positioning method based on a quantization window according to claim 4, characterized in that: The correction of abnormal sub-windows in spatial parallel constraint matching relationships specifically includes: The system obtains the relative position transformation vector between keyframe pairs based on the starting keyframe in the current quantization window and the matching keyframe of that keyframe in the reference quantization window. This allows us to infer the corrected position vector of the current keyframe that triggered the correction within the current quantization window. : in, This represents the position vector of the current keyframe matched within the reference quantization window; since the quantization window correction algorithm focuses on eliminating the cumulative error of robot yaw angle estimation, the algorithm uses the original keyframe position vector... The component of gravity in the direction of gravity is retained, and the final vector is derived. ; Taking the coordinate system of the starting keyframe in the current quantization window as the principal coordinate system, we obtain... and Unit vector in the principal coordinate system: In the formula: Represents the principal coordinate system; in formula (7), using and Solve for the axis angles required to correct the principal coordinate system. : pass Further, the pose transformation matrix required for correction is obtained. : In formula (8), the exponential mapping is expressed as ,in It is the rotation matrix required for quantization window correction; using right All keyframe poses are corrected and propagated; the pose transformation of each corrected keyframe relative to the world coordinate system is performed. The calculation is as follows: The transformation relationship between the velocity vectors of each keyframe and the principal coordinate system remains unchanged before and after correction. Obtain the corrected velocity vectors for each keyframe in the world coordinate system. : After the keyframes are corrected, the position of the point cloud is corrected according to the transformation relationship between the keyframes and the map point cloud.

6. A stereo vision inertial odometry positioning system based on a quantization window, characterized in that, The system includes: The receiving module is used to acquire the latest keyframe obtained by the visual inertial odometry through preliminary calculation and estimation; The quantization module is used to input the latest keyframe into the preset quantization window subsystem, quantify the robot's latest driving state at the keyframe level, and obtain the corresponding driving state quantization index. The separation module is used to continuously separate and construct the current quantization window from the robot's historical trajectory based on the quantization index of the driving state, and selectively construct a reference quantization window based on the constructed current quantization window during the current quantization window reset phase. The parallel constraint module is used to construct spatial parallel constraints between the current quantization window and the reference quantization window. The detection and correction module is used to detect anomalies in spatial parallel constraint matching relationships, correct the sub-windows of anomalies in spatial parallel constraint matching relationships, and complete map updates. The step of inputting the latest keyframe into a preset quantization window subsystem to quantize the robot's latest driving state at the keyframe level and obtain the corresponding driving state quantization index specifically includes: In the pre-defined quantization window subsystem, maintain a window of length... The most recent keyframe queue; simulation based on the structural characteristics of crop rows in the field. The theoretical location of the nearest keyframe, where, This represents the length of the temporary window in the current quantization window; the reverse recursive formula for the theoretical position of each most recent keyframe in the most recent keyframe queue is as follows: in, and These represent the theoretical and estimated positions of the most recent keyframe, respectively. Represents the first two dimensions of the matrix. Indicates the horizontal distance between two locations; The robot's driving state at the moment of the latest keyframe is quantified by calculating the horizontal trajectory error between the actual keyframe sequence and the simulated keyframe sequence; the formula for calculating the driving state quantization index is as follows: in, and These represent the actual estimated pose of the system and the simulated theoretical pose, respectively. This represents the projection of the translation vector of the pose matrix onto the XY plane; It represents the square of the Euclidean norm of a vector.

7. A computer-readable storage medium for storing one or more programs, characterized in that, The one or more programs include instructions that, when executed by a computing device, cause the computing device to perform any of the methods according to claims 1-5.

8. A computing device, characterized in that, include: One or more processors, a memory, and one or more programs, wherein the one or more programs are stored in the memory and configured to be executed by the one or more processors, the one or more programs including instructions for performing any of the methods according to claims 1-5.

Images