Laser SLAM degradation processing method based on dual threshold detection and adaptive compensation

Abstract

The invention discloses a laser SLAM (Simultaneous Localization and Mapping) degradation treatment method based on dual threshold detection and adaptive compensation, which is applied to the technical field of degradation detection. Comprising the following steps: constructing a dual detection mechanism, ensuring the detection stability in a conventional scene by a static threshold value, adjusting a judgment standard by a dynamic threshold value, and realizing rapid identification of progressive and instantaneous degradation; designing a self-adaptive two-stage compensation strategy driven by degradation deviation, wherein first-stage compensation is based on a vector projection principle; and the secondary compensation dynamically weakens the influence of unreliable laser constraint according to the degradation deviation degree. According to the method, the severity of degradation deviation is quantified in real time based on a dual-threshold detection mechanism of scene feature real-time perception, starting and intensity adjustment of a compensation strategy are driven, and the dual detection mechanism and a two-stage compensation strategy are subjected to linkage design, so that the scene adaptability and detection comprehensiveness can be improved, the compensation intensity can be dynamically adjusted, and the detection accuracy is improved. And the pose estimation precision and robustness of the system are obviously improved.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of degradation detection, and more particularly to a laser SLAM degradation processing method based on double threshold detection and adaptive compensation. BACKGROUND

[0002] With the rapid development of artificial intelligence and automation technology, the field of positioning and navigation is also constantly undergoing innovation. Simultaneous Localization and Mapping (SLAM) as a key technology to endow mobile robots with environmental perception and autonomous navigation capabilities, can enable the carrier to construct a map based on the information increment collected by sensors under the condition of unknown initial pose and environment, and simultaneously determine its real-time position in the map. Laser-based SLAM technology is widely used due to its fast and accurate acquisition of environmental information, strong anti-interference, and immunity to changes in light, etc. However, in the scenes such as coal mine tunnels, indoor corridors, and tunnels with monotonous geometric structure, sparse texture features, or high repetition, due to the limited scanning range of laser radar and the strong repetition of environmental features, the distribution of point clouds is different, causing constraint imbalance. This constraint loss causes significant cumulative drift in a particular direction of pose estimation, and as time goes on, the positioning error continues to expand, eventually leading to a sharp decline in positioning accuracy and the system falling into a degradation dilemma. To overcome the positioning problem of laser SLAM algorithm in degradation scenarios, multi-source sensor fusion technology has gradually become the focus of research, among which tight coupling LIO that puts laser radar LiDAR and inertial measurement unit (IMU) together for joint processing, considering the internal relationship and mutual influence has become the mainstream direction.

[0003] Existing mainstream tight coupling methods (such as LIO-SAM and FAST-LIO) usually adopt a static or slowly adjusted weight distribution strategy to globally fuse the measurement residuals of laser radar and IMU. However, SLAM degradation is often instantaneous, and when the constraints of laser in a particular direction are temporarily invalid, the system cannot respond quickly, neither can it timely reduce the weight of unreliable laser constraints, nor can it simultaneously increase the compensation weight of IMU in that direction, leading to continuous pose estimation deviation caused by unreliable geometric constraints, and eventually causing trajectory deviation.

[0004] In the degradation detection link, the existing technology mostly relies on a single fixed threshold method, which only sets a fixed numerical threshold. If the eigenvalue of the information matrix in the laser SLAM optimization process is lower than the threshold, it is determined that degradation occurs. This detection method has obvious limitations, which directly leads to a serious lack of detection accuracy. On the one hand, the threshold setting is highly dependent on human experience and specific scenarios, and it is difficult to adapt to the normal fluctuation range of the eigenvalue in different environments, and it is easy to have a "threshold mismatch" problem in complex and variable actual scenarios. On the other hand, the fixed threshold only focuses on the static value of the eigenvalue, ignoring the key information contained in the dynamic changes of the eigenvalue. For gradual degradation of "slowly decreasing eigenvalue" (such as a robot gradually entering a long corridor from a feature-rich area), the fixed threshold often triggers detection when the degradation is already serious, missing the early intervention opportunity. For instantaneous degradation of "eigenvalue with severe fluctuations or cliff-like decline" (such as a laser radar being suddenly blocked), it may also cause misjudgment due to the threshold not covering the extreme fluctuation range, and cannot timely identify the sharp deterioration of the observation data quality. In summary, a single fixed threshold detection cannot achieve comprehensive and accurate perception of different types of degradation, which has become one of the core bottlenecks restricting system reliability. Moreover, the current field lacks targeted compensation research based on the degradation detection results. Even if the existing scheme completes the degradation detection, it does not establish a correlation mechanism between the detection results and the compensation strategy, and is still limited to simply fusing IMU and SLAM data through a fixed weight, which cannot adjust the compensation strategy according to the direction and degree of degradation. Such indiscriminate compensation not only fails to alleviate the pose estimation bias caused by degradation, but also may introduce new errors due to improper compensation, further reducing system performance. There are still the following problems in the degradation detection of the long corridor: the degradation detection is lagging and inaccurate, relies on fixed threshold judgment, only focuses on the static value of the eigenvalue, and cannot capture the gradual degradation of the slowly decreasing eigenvalue, and is easy to miss detection and misjudgment due to poor threshold adaptability; it cannot locate the specific direction of degradation, and lacks effective compensation means matching the detection results, making it difficult to solve the pose estimation bias in a targeted manner. Therefore, how to provide a technical solution with "accurate degradation detection" and adaptive directional compensation" capabilities is a problem that needs to be solved by those skilled in the art. SUMMARY

[0005] Therefore, the present application provides a laser SLAM degradation processing method based on double threshold detection and adaptive compensation, a double threshold detection mechanism based on real-time perception of scene features, real-time quantification of the severity of degradation bias, and driving of the start and intensity adjustment of the compensation strategy. The double detection mechanism and two-level compensation strategy are designed in linkage to solve the problems existing in the degradation scenarios such as long corridors and tunnels.

[0006] In order to achieve the above purpose, the present application provides the following technical scheme: A laser SLAM degradation processing method based on dual threshold detection and adaptive compensation includes the following steps: S1. Perform eigenvalue decomposition on the Hessian matrix to obtain eigenvalues ​​and eigenvectors. Use the magnitude of the eigenvalues ​​as a quantitative indicator of the constraint strength of each degree of freedom, and use the direction of the eigenvectors as a criterion for the determination of the degradation direction. S2. Design a dual threshold detection mechanism based on real-time scene feature perception. Perform statistical analysis on the set of eigenvalues ​​of the Hessian matrix to calculate the static threshold. Use a sliding window to store the eigenvalue sequence, perform statistical analysis on the eigenvalue sequence within the sliding window, and calculate the dynamic threshold. S3. A dual-threshold detection mechanism based on real-time scene feature perception quantifies the severity of degradation deviation in real time and calculates the deviation rate by combining static and dynamic thresholds. Laser SLAM degradation direction is detected based on static and dynamic thresholds, and based on the deviation rate. Assess the degree of degradation; S4. The compensation strategy is initiated and its intensity is adjusted based on a dual threshold detection mechanism. This is done according to the degradation direction and degree detected in S2, and based on the deviation rate. Determine the degree of degradation and initiate a two-stage progressive compensation strategy, linking the dual detection mechanism with the two-stage compensation strategy, including: When the deviation rate δ < 0.2, it is judged as slight degradation, and the first-level compensation based on the vector projection principle is initiated; When the deviation rate δ≥0.2, it is judged as moderate or above degradation, and secondary compensation based on IMU adaptive weighted fusion is initiated.

[0007] Optionally, the static threshold detection in S2 specifically includes: Statistical analysis was performed on the eigenvalue set of the Hessian matrix, and the minimum value of the normal eigenvalues ​​was selected. Multiply by safety factor To obtain the static threshold : ; Since different degrees of freedom of motion have different sensitivities to constraints, a static threshold needs to be set independently for each degree of freedom. The feature values ​​obtained by optimizing the current frame and Compare, if If so, the system initially determines that there is degradation in that direction.

[0008] Optionally, the dynamic threshold detection in S2 specifically includes: A sliding window of fixed size is used to store the recent eigenvalue sequence. , the window update rule is: when the data amount in the window does not reach SIZE, the new feature value is directly entered into the window, and the historical data is all retained; when the data amount in the window reaches SIZE, the new feature value is entered into the window at the same time, and the earliest feature value entered into the window is exited from the window, so that the data amount in the window is maintained constant; The statistical analysis is performed on the feature value sequence in the sliding window, and the first quartile and the third quartile of the feature value sequence are calculated; A dynamic threshold is set as the lower limit of the normal range of the feature value, and the formula is: ; In the formula, the hyperparameter controls the detection sensitivity, if the current frame feature value satisfies , it is determined that the dynamic degradation exists in the direction.

[0009] Optionally, the detection of the degradation direction and the degradation degree of the laser SLAM in S3 is specifically: A bias rate is introduced as a core quantitative index of the degradation degree, and the bias rate directly reflects the credibility of the laser odometry; in order to quantify the degradation degree, a static bias value and a dynamic bias value are defined: ; ; Comprehensive static and dynamic detection results are provided, and a unified degradation degree quantitative index is defined, and a comprehensive bias value is defined as the bias rate : ; In the formula, is a scene coefficient.

[0010] Optionally, the first-level compensation in S4 is specifically: According to the accurate recognition of the degradation direction based on the double threshold mechanism, the constraint vector in the degradation direction is projected to other motion dimensions without degradation, the remaining effective constraint part is used to offset the influence caused by the degradation, and the error component in the original degradation direction is projected to the reliable dimension, so that the error is effectively suppressed.

[0011] Optionally, the second-level compensation in S4 is specifically: The high-frequency pose data obtained by IMU integration is fused with the unreliable pose estimation of the laser odometry in the degradation direction, the bias rate δ output by the double threshold mechanism is used to dynamically adjust the fusion weight, the fusion effect is optimized, and the fusion formula is: ; In the formula, is the compensated pose, is the initial pose provided by laser SLAM, is the pose calculated by IMU integration, is the weight coefficient adaptively adjusted according to the degradation degree.

[0012] Optionally, the weight coefficient is dynamically calculated according to the deviation rate δ: ; Wherein: is the deviation rate of the current frame degradation degree; is the minimum value of the IMU weight, used for no degradation or slight degradation scene; is the maximum value of the IMU weight, used for limiting the maximum weight when the degradation is serious, to avoid the excessive dominance of IMU.

[0013] Through the above technical solution, compared with the prior art, the present application provides a laser SLAM degradation processing method based on double threshold detection and adaptive compensation, which has the following beneficial effects: the present application can cover two types of degradation, gradual and instantaneous, through the double threshold detection mechanism based on real-time scene feature perception, can accurately identify the degradation state in real time, and accurately locate the degradation direction and quantify the degradation degree, solving the problems of traditional detection lag, misjudgment and unknown direction; the deviation rate δ output by the double threshold detection is used to dynamically adjust the fusion weight, the more serious the degradation degree (the larger the deviation rate δ), the higher the IMU weight ratio is increased to strengthen its pose constraint effect, and the laser odometry weight is reduced to reduce the interference of unreliable information; otherwise, the laser odometry contribution is moderately retained; through adaptive weighted fusion, directional correction of laser odometry output is realized, effectively suppressing the positioning drift, ensuring that the pose estimation accuracy is still high in serious degradation scene; through the start and intensity adjustment of the compensation strategy driven by double detection, the double detection mechanism and two-stage compensation strategy are designed in linkage, based on the deep cooperation of double detection and adaptive compensation, the full scene coverage of long corridor, tunnel and other degradation scenes can be realized, the detection accuracy and compensation effectiveness of the laser SLAM system are significantly improved, and the robustness of the system in complex environment is greatly enhanced. BRIEF DESCRIPTION OF DRAWINGS

[0014] In order to more clearly illustrate the technical solutions in the embodiments of the present application or the prior art, the following will briefly introduce the drawings needed to be used in the embodiment or prior art description. Obviously, the drawings in the following description are only embodiments of the present application, and for those skilled in the art, other drawings can be obtained without creative labor on the basis of the provided drawings.

[0015] Figure 1A flow chart of the laser SLAM degradation processing method based on dual threshold detection and adaptive compensation of the present application is shown in Figure 1. Figure 2 A flow chart of the dual degradation detection mechanism of the present application is shown in Figure 2. Figure 3 A schematic diagram of the dual degradation detection result in the embodiment of the present application is shown in Figure 3. Figure 4 A schematic diagram of the absolute error comparison in the embodiment of the present application is shown in Figure 4. DETAILED DESCRIPTION

[0016] The technical solutions in the embodiments of the present application will be described clearly and completely below with reference to the accompanying drawings in the embodiments of the present application. Obviously, the described embodiments are only a part of the embodiments of the present application, rather than all the embodiments. Based on the embodiments in the present application, all other embodiments obtained by those skilled in the art without creative work fall within the scope of protection of the present application.

[0017] The embodiment of the present application discloses a laser SLAM degradation processing method based on dual threshold detection and adaptive compensation, as shown in Figure 1, which comprises the following steps: Figure 1 S1, performing eigenvalue decomposition on the Hessian matrix to obtain eigenvalues and eigenvectors, taking the eigenvalue size as a quantitative index of the constraint strength of each degree of freedom, and taking the eigenvector direction as a discrimination basis of the degradation direction; S2, designing a dual threshold detection mechanism based on real-time perception of scene features, performing statistical analysis on the eigenvalue set of the Hessian matrix to calculate a static threshold, and using a sliding window to store the eigenvalue sequence to perform statistical analysis on the eigenvalue sequence in the sliding window to calculate a dynamic threshold; the dual threshold detection mechanism is shown in Figure 2; Figure 2 S3, based on the dual threshold detection mechanism of real-time perception of scene features, quantifying the severity of the degradation deviation in real time, calculating the deviation rate based on the static threshold and the dynamic threshold , detecting the laser SLAM degradation direction based on the static threshold and the dynamic threshold, and judging the degradation degree based on the deviation rate S4, driving the start and intensity adjustment of the compensation strategy based on the dual threshold detection mechanism, determining the degradation degree according to the deviation rate according to the degradation direction and the degradation degree detected in S2, linking the dual detection mechanism with the two-stage compensation strategy, including: When the deviation rate δ < 0.2, it is determined as slight degradation, and a first-stage compensation based on the vector projection principle is started; When the deviation rate δ ≥ 0.2, it is determined as moderate and above degradation, and a second-stage compensation based on the adaptive weighted fusion of IMU is started.​​​

[0018] In this embodiment of the invention, S1 specifically refers to: Calculate the Hessian matrix: ; ; In the formula, It is the n×6 Jacobian matrix of the residual equation with respect to the optimization variables. Each row corresponds to the partial derivative of the residual function with respect to the six pose degrees of freedom (roll, pitch, yaw, x-axis translation, y-axis translation, z-axis translation). Let n be the residual vector. It is a 6-dimensional pose increment vector; Eigenvalue decomposition of the Hessian matrix is ​​performed, and the magnitude of the eigenvalues ​​directly quantifies the reliability of the pose solution at the corresponding direction or rotation angle. By analyzing the eigenvalues ​​and eigenvectors of the Hessian matrix, a basis for degradation detection and compensation is provided. Specifically: By using the Jacobian rotation method, a series of orthogonal rotation matrices are iteratively constructed to gradually eliminate the off-diagonal elements of the Hessian matrix, ultimately transforming the Hessian matrix into a diagonal matrix and obtaining the corresponding eigenvector matrix. The specific decomposition results satisfy the following: ; In the formula, These are orthogonal eigenvectors. yes The i 1 eigenvector It is a diagonal matrix. yes The i One eigenvalue; The magnitude of eigenvalues ​​reflects the constraint strength of a given degree of freedom. In feature-rich, non-degenerate environments, point clouds provide strong constraints across all degrees of freedom, resulting in a full-rank Hessian matrix. When degradation occurs, the Hessian matrix exhibits singular or near-singular characteristics in the corresponding degree of freedom dimension. Specifically, the eigenvalue corresponding to that degree of freedom is significantly smaller than other eigenvalues, directly reflecting a sharp decrease in constraint strength in that direction. Therefore, the eigenvalues ​​and eigenvectors of the Hessian matrix have a clear mathematical relationship with constraints in laser SLAM systems: the magnitude of the eigenvalue directly maps to the strength of the constraint, while the direction of the eigenvector indicates the specific motion dimension affected by the constraint. When a certain eigenvalue decreases significantly, its corresponding eigenvector can accurately identify the degree of freedom where the constraint is missing. Based on this characteristic, analyzing the eigenvalues ​​and eigenvectors of the Hessian matrix provides a more scientific and dynamic basis for degradation detection and compensation.

[0019] Furthermore, the static threshold detection in S2 specifically involves: The core objective of static threshold detection is to quickly filter out degraded frames with severely lacking constraints. Its technical principle is based on the quantitative relationship between the eigenvalues ​​of the Hessian matrix and the constraint strength: when the eigenvalue corresponding to a certain degree of freedom of motion is too small, it indicates that the point cloud features in that direction cannot provide effective constraints, resulting in low pose calculation reliability, i.e., a risk of degradation. The static threshold needs to be determined through offline statistics to ensure that the eigenvalue is consistently higher than the threshold under normal operating conditions and significantly lower than the threshold during severe degradation. Statistical analysis was performed on the eigenvalue set of the Hessian matrix, and the minimum value of the normal eigenvalues ​​was selected. Multiply by safety factor To obtain the static threshold : ; Since different degrees of freedom of motion have different sensitivities to constraints, a static threshold needs to be set independently for each degree of freedom. The feature values ​​obtained by optimizing the current frame and Compare, if If so, the system initially determines that there is degradation in that direction.

[0020] In this embodiment of the invention, under non-degradable conditions (such as structured environments or regions rich in point cloud features), pose calculation results of no less than 500 frames are collected, and the Hessian matrix eigenvalues ​​corresponding to each degree of freedom of motion in each frame are extracted to form a normal eigenvalue set. Among them, the safety factor The value ranges from 0.5 to 0.8. The stronger the scene interference, the better. The smaller the value, the lower the probability of misjudgment caused by normal fluctuations.

[0021] Furthermore, the dynamic threshold detection in S2 specifically involves: To address the issues of static thresholding failing to capture gradual degradation and being prone to misjudgment, an interquartile range (IQR) algorithm based on a sliding window is introduced to construct an anomaly detection model for dynamically changing feature values. This model adaptively defines a "normal range" by analyzing the statistical distribution of historical feature value sequences, focusing on identifying both gradual degradation with slowly decreasing feature values ​​and instantaneous interference from sharp fluctuations, thus compensating for the scene adaptability limitations of static thresholding. A sliding window of fixed size is used to store the recent eigenvalue sequence. The window updating rule is: when the data amount in the window does not reach SIZE, the new feature value is directly entered into the window, and all the historical data are retained; when the data amount in the window reaches SIZE, the new feature value is entered into the window at the same time, and the earliest feature value entered into the window is exited from the window, so that the data amount in the window is kept constant; The first quartile (the 25% quantile of the sequence) and the third quartile (the 75% quantile of the sequence) of the feature value sequence in the sliding window are calculated, and a dynamic threshold is set as the lower limit of the normal range of the feature value, and the formula is: ; In the formula, the hyperparameter controls the detection sensitivity, if the current frame feature value satisfies , it is determined that the dynamic degradation exists in the direction.

[0022] Further, the detection of the degradation direction and the degradation degree of the laser SLAM in S3 is specifically: A double detection mechanism combining a static threshold and a sliding window IQR algorithm is adopted to comprehensively determine the degradation event. In the mechanism, the static threshold module is used for rapid response to significant degradation; and the IQR algorithm is used for adaptively capturing the gradual or instantaneous degradation trend by analyzing the historical distribution of the feature value. The two work together to jointly guarantee the detection robustness and timeliness of the system in various degradation scenarios; The bias rate is introduced as a core quantitative index of the degradation degree, and the bias rate directly reflects the credibility of the laser odometer; in order to quantify the degradation degree of detection, the static bias value and the dynamic bias value are defined: ; ; Comprehensive static and dynamic detection results are provided to provide a unified degradation degree quantitative index, and the comprehensive bias value is defined as the bias rate : ; In the formula, is a scene coefficient.

[0023] In the embodiment of the application, when the environment is stable, 0.2-0.4 are taken, dynamic detection is emphasized to adapt to gradual degradation; when the environment changes greatly or sudden interference occurs, 0.6-0.8 are taken, static detection is emphasized to quickly respond to significant degradation; By effective point cloud ratio determine whether the environment has a dramatic change or a sudden disturbance, the effective point cloud ratio is the core quantitative index of laser odometry observation data quality, directly reflects the scene structure integrity and sensor data effectiveness, and provides an objective basis for dynamic adjustment: ; In the formula, is the number of effective point clouds, is the total number of point clouds in the current frame; under normal circumstances, usually stabilizes at 0.6~0.9 (usually > 0.8 for indoor structured scenes, and usually 0.6~0.7 for outdoor cluttered scenes); if , it can be determined that the environment has a dramatic change (such as entering an outdoor cluttered scene from an indoor structured scene) or a sudden disturbance (such as a sharp reduction in effective points due to contamination of the sensor lens).

[0024] Further, the first compensation in S4 is specifically: According to the accurate recognition of the degradation direction based on the Hessian matrix eigenvalue analysis, the constraint vector in the degradation direction is projected to other motion dimensions that do not occur degradation, and the remaining effective constraint part is used to offset the influence of degradation. By projecting the error component in the original degradation direction to the reliable dimension, the error is effectively suppressed.

[0025] Specifically: first, extract the original constraint vector in the degradation direction from the 6×6 Hessian matrix H: ; The reliable direction matrix is , the orthogonal projection matrix is constructed to ensure that only the reliable dimension is projected, because direct projection is zero, and reliable observations (such as ground matching points) are extracted to reconstruct the constraint to full rank, and compensate for the loss of constraints.

[0026] Further, the second compensation in S4 is specifically: fuse the high-frequency pose data obtained by IMU integration with the unreliable pose estimation of laser odometry in the degradation direction, and dynamically adjust the fusion weight using the deviation rate δ output by the double threshold mechanism to optimize the fusion effect. The fusion formula is: ; In the formula, is the compensated pose, is the initial pose provided by laser SLAM, is the pose calculated by IMU integration, is the weight coefficient adaptively adjusted according to the degradation degree.

[0027] Further, the weight coefficient is dynamically calculated according to the deviation rate δ: ; Wherein: is the deviation rate of the degradation degree of the current frame; is the minimum value of the IMU weight, used for no degradation or slight degradation scenarios; is the maximum value of the IMU weight, used for limiting the maximum weight when the degradation is serious, so as to avoid the over-dominance of the IMU.

[0028] In the embodiment of the present application, the value is 0.1, the value is 0.9; the larger the δ is, the lower the reliability of the laser odometer is, and the IMU weight is increased accordingly, so that the IMU data is dominant in the case of serious degradation, and the robustness of the pose estimation is improved. The compensation strategy takes the deviation rate δ as the quantitative basis to realize on-demand compensation and avoid early or late start; the compensation process is adaptive: the IMU weighting coefficient is dynamically adjusted according to the degradation degree, and the adaptability to the dynamic characteristics of complex scenes is improved.

[0029] In an embodiment of the present application, an experimental environment is built on a VMware virtual machine platform, and an indoor long corridor data set with a total length of 616.45 meters is selected for testing. The data set contains a back-and-forth motion trajectory, in which there are multiple corridor sections with sparse feature points and prone to degradation, which is suitable for verifying the effectiveness of the degradation scene algorithm. The FAST-LIO2 algorithm, the LIO-SAM algorithm and the algorithm of the embodiment of the present application are compared and tested, and mapping and trajectory estimation are performed based on the same data set.

[0030] The double degradation detection results are as shown in Figure 3 Due to the introduction of the IQR algorithm, the system can effectively identify abnormal situations where the feature value sequence changes significantly (such changes may be caused by dynamic object interference or increased sensor noise), even if the value is much higher than the static threshold, it can still be accurately detected. The absolute error comparison chart is as shown in Figure 4 The experimental results show that the absolute error generated by the algorithm of the present application in the entire running process is less than that of the LIO-SAM and FAST-LIO2 algorithms, showing higher estimation accuracy. At the same time, the deviation fluctuation amplitude is significantly reduced, indicating that the pose solving result is more stable, and the fitting degree of the estimated trajectory and the real path is higher.

[0031] The various embodiments described herein are presented by way of example only and are not intended as a limitation on the scope of the disclosure. As those skilled in the art will understand, the principles and operation of the present disclosure can be applied using any number of other embodiments as well and each of the various embodiments presented herein can be implemented in a variety of ways.

[0032] Various modifications to these embodiments will be readily apparent to those skilled in the art, and the generic principles defined herein can be applied to other embodiments without departing from the spirit or scope of the disclosure. Thus, the present disclosure is not intended to be limited to the embodiments shown herein but is to be accorded the widest scope consistent with the principles and novel features disclosed herein.

Claims

1. A laser SLAM degradation processing method based on dual threshold detection and adaptive compensation, characterized in that, Includes the following steps: S1. Perform eigenvalue decomposition on the Hessian matrix to obtain eigenvalues ​​and eigenvectors. Use the magnitude of the eigenvalues ​​as a quantitative indicator of the constraint strength of each degree of freedom, and use the direction of the eigenvectors as a criterion for the determination of the degradation direction. S2. Design a dual threshold detection mechanism based on real-time perception of scene features, perform statistical analysis on the set of Hessian matrix eigenvalues, and calculate the static threshold. A sliding window is used to store the feature value sequence. Statistical analysis of the feature value sequence within the sliding window is performed to calculate the dynamic threshold. S3. A dual-threshold detection mechanism based on real-time scene feature perception quantifies the severity of degradation deviation in real time and calculates the deviation rate by combining static and dynamic thresholds. Laser SLAM degradation direction is detected based on static and dynamic thresholds, and based on the deviation rate. Assess the degree of degradation; S4. The compensation strategy is initiated and its intensity is adjusted based on a dual threshold detection mechanism. This is done according to the degradation direction and degree detected in S2, and based on the deviation rate. Determine the degree of degradation and initiate a two-stage progressive compensation strategy, linking the dual detection mechanism with the two-stage compensation strategy, including: When the deviation rate δ < 0.2, it is judged as slight degradation, and the first-level compensation based on the vector projection principle is initiated; When the deviation rate δ≥0.2, it is judged as moderate or above degradation, and secondary compensation based on IMU adaptive weighted fusion is initiated.

2. The laser SLAM degradation processing method based on dual threshold detection and adaptive compensation according to claim 1, characterized in that, The static threshold detection in S2 is specifically as follows: Statistical analysis was performed on the eigenvalue set of the Hessian matrix, and the minimum value of the normal eigenvalues ​​was selected. Multiply by safety factor To obtain the static threshold : ； Since different degrees of freedom of motion have different sensitivities to constraints, a static threshold needs to be set independently for each degree of freedom. The feature values ​​obtained by optimizing the current frame and Compare, if If so, the system initially determines that there is degradation in that direction.

3. The laser SLAM degradation processing method based on dual threshold detection and adaptive compensation according to claim 1, characterized in that, The dynamic threshold detection in S2 is specifically as follows: A sliding window of fixed size is used to store the recent eigenvalue sequence. The window update rule is as follows: when the amount of data in the window is less than SIZE, new feature values ​​are directly added to the window, and all historical data are retained; when the amount of data in the window reaches SIZE, new feature values ​​are added to the window, and the earliest added feature values ​​are removed from the window, thus maintaining a constant amount of data in the window. Perform statistical analysis on the sequence of eigenvalues ​​within the sliding window and calculate its first quartile. and the third and fourth quartiles Set dynamic threshold As the lower bound of the normal range of eigenvalues, the formula is: ； In the formula, hyperparameters Control the detection sensitivity if the current frame feature value satisfy If so, it is determined that there is dynamic degradation in that direction.

4. The laser SLAM degradation processing method based on dual threshold detection and adaptive compensation according to claim 1, characterized in that, The specific steps for detecting the degradation direction and degree of laser SLAM in S3 are as follows: Introducing deviation rate As a core quantitative indicator of the degree of degradation, the deviation rate It directly reflects the reliability of the laser odometer; to quantify the degree of degradation in the detection, a static deviation value is defined. Dynamic deviation value : ； ； By combining static and dynamic detection results, a unified quantitative index for the degree of degradation is provided, and the comprehensive deviation value is defined as the deviation rate. : ； In the formula, For scene coefficients.

5. The laser SLAM degradation processing method based on dual threshold detection and adaptive compensation according to claim 1, characterized in that, The first-level compensation in S4 is as follows: Based on the degradation direction accurately identified by the dual threshold mechanism, the constraint vector in the degradation direction is projected to other motion dimensions that have not undergone degradation. The remaining effective constraint part is used to offset the impact of degradation. By projecting the error component in the original degradation direction to the reliable dimension, the error is effectively suppressed.

6. The laser SLAM degradation processing method based on dual threshold detection and adaptive compensation according to claim 1, characterized in that, The secondary compensation in S4 is specifically as follows: The high-frequency pose data obtained by IMU integration is fused with the unreliable pose estimation of the laser odometry in the degradation direction. The deviation rate δ output by the dual threshold mechanism is used to dynamically adjust the fusion weights to optimize the fusion effect. The fusion formula is as follows: ； In the formula, It is the compensated pose. The initial pose is provided by laser SLAM. It is the pose calculated by IMU integration. It is a weighting coefficient that is adaptively adjusted according to the degree of degradation.

7. The laser SLAM degradation processing method based on dual threshold detection and adaptive compensation according to claim 6, characterized in that, Weighting coefficient Dynamically calculated based on the deviation rate δ: ； in: The degradation deviation rate of the current frame; This represents the minimum value of the IMU weights, used for scenarios with no or slight degradation. This is the maximum value of the IMU weights, used to limit the maximum weights in cases of severe degradation, preventing the IMU from becoming overly dominant.

Images